Philosophical Grammar

LUDWIG WITTGENSTEIN

PHILOSOPHI CAL GRAMMAR PART I

The Proposition and its Sense PART II

On Logic and Mathematics

Edited by

RUSH RHEES Translated by

ANTHONY KENNY I

,

--.--~,.,,-

BASIL BLACKWELL

.

OXFORD

©

Basil Blackwell, Oxford, 1974

CONTENTS

All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of Basil Blackwell & Mott Limited.

Part I The Proposition and its Sense

I 1 How can one talk about "understanding" and "not understanding" a proposition? Surely it's not a proposition until it's understood? 39

iilii"iljiiiijj

2 Understanding and signs. Frege against the formalists. Understanding like seeing a picture that makes all the rules clear; in that case the picture is itself a sign, a cal~ulus. "To understand a language" - to take in a symbolism as a whole. Language must speak for itself. 39

380501050055022

3 One can say that meaning drops out of language. In contrast: "Did you mean that seriously or as a joke?" When we mean (and don't just say) words it seems to us as if there were something coupled to the words. 41

SIIR. JlOr

Jsr:

4 Comparison with understanding a piece of music: for explanation I can only translate the musical picture into a picture in another medium - and why just that picture? Comparison with understanding a picture. Perhaps we see only patches and lines "we do not understand the picture". Seeing a genre-picture in different ways. 41

ISBN 0631 152202

5 "I understand that gesture" - it says something. In a sentence a word can be felt as belonging first with one word and then with another. A 'proposition' may be what is conceived in different ways or the way of conceiving itself.

Printed in Great Britain by William Clowes & Sons Limited London, Colchester and Beccles and Bound at the Kemp Hall Bindery, Oxford

I

-..

A sentence from the middle of a story I have not read. The concept of understanding is a fluid one.

It is not a question of an instantaneous grasping. When a man who knows the game watches a game of chess, the experience he has when a move is made usually differs from that of someone else watching without understanding the game. But this experience is not the knowledge of the rules. - The understanding of language seems like a background; like the ability to multiply. 11

42

6 A sentence in a code: at what moment of translating does understanding begin? The words of a sentence are arbitrary; so I replace them with letters. But now I cannot immediately think the sense of the sentence in the new expression. The notion that we can only imperfectly exhibit our understanding: the expression of understanding has something missing that is essentiallY inexpressible. But in that case it makes no sense to speak of a more complete expression. 43 7 What is the criterion for an expression's being meant thus? A question about the relationship between two linguistic expressions. Sometimes a translation into another mode of representation. 45 8 Must I understand a sentence to be able to act on it? If "to

understand a sentence" means somehow or other to act on it, then understanding cannot be a precondition for our acting on it. What goes on when I suddenly understand someone else? There are matry possibilities here. 45

12 When do we understand a sentence? - When we've uttered the whole of it? Or while uttering it? 50

13 When someone interprets, or understands, a sign in one sense or another, what he is doing is taking a step in a calculus."Thought" sometimes means a process which may accompany the utterance of a sentence and sometimes the sentence itself in the 50 system of language.

II 14 Grammar as (e.g.) the geometry of negation. We would like to say: "Negation has the property that when it is doubled it yields an affirmation". But the rule doesn't give a further description of negation, it constitutes negation. 52

9 Isn't there a gap between an order and its execution? "I understand it, but only because I add something to it, namely the interpretation." - But if one were to say "any sentence still stands in need of an interpretation", that would mean: no sentence can be unde:r:stood without a rider. 46

15 Geometry no more speaks about cubes than logic does about negation. It looks as if one could infer from the meaning of negation that "~~ p" means p. 52

"Understanding a word" - being able to apply it. - "When I sain ---_ -+

A

The resulting river takes the name of that source in whose approximate direction it flows onward.) Think of the possible criteria of identity for things like colour patches in my visual field (or figures on a cinema screen) and of the different kinds of use of names given to such patches or figures. If we turn to the form of expression "(3x). fx" it's clear that this is a sublimation of the form of expression in our language: "There are human beings on this island" "There are stars that we do not see". To every proposition of the form "(3x). fx" there is supposed to correspond a proposition "fa", and "a" is supposed to be a name. So one must be able to say "(3x). fx, namely a and b",

1. There appears to be something wrong with the German text here. Possibly Wittgenstein meant to write "let this man be called 'N'" and inadvertently wrote a version which is the same as the one he is correcting. (Trs.)

202

l~

("There are some values of x, which satisfy fx, namely a and b"), or "(3x). fx, e.g. a", etc. And this is indeed possible in a case like "There are human beings on this island namely Messrs ABC D." But then is it essential to the sense of the sentence "Th~re' ar~ men on this island" that we should be able to name them, and fix a particular criterion for their identification? That is only so in the case where the proposition "(3x). fx" is defined as a disjunction of propositions of the form "f(x)", if e.g. it is laid down that "There are men on this island" means "Either Mr. A or Mr. B or Mr. C or Mr. D. or Mr. E is on this island" - if, that is, one determines the concept "man" extensionally (which of course is quite contrary to the normal use of this word.) (On the other hand the concept "primary colour" really is determined extensionally.) So it doesn't always make sense when presented with a proposition "(3x). fx" to ask "Which xs satisfy f?" "Which red circle a centimetre across is in the middle of this square"? - One mustn't confuse the question "which object satisfies f?" with the question "what sort of object ... etc. ?" The first question would have to be answered by a name, and so the answer would have to be able to take the form "f(a)"; the question "what sort of . .. ?" is answered by "(3x). fx. q>x". So it may be senseless to ask "which red spot do you see ?" and yet make sense to ask "what kind of a red spot do you see (a round one, a square one, etc.) ?" I would like to say: the old logic contains more convention and physics than has been realised. If a noun is the name of a boc(y, a verb is to denote a movement, and an adjective to denote a property of a body, it is easy to see how much that logic presupposes; and it is reasonable to conjecture that those original presuppositions go still deeper into the application of the words, and the logic of propositions. (Suppose we were set the task of projecting figures of various shapes on a given plane I into a plane II. We could then fix a

method of projection (say orthogonal projection) and carry out the mapping in accordance with it. We could also easily make inferences from the representations on plane II about the figures on plane 1. But we could also adopt another procedure: we might decide that the representations in the second plane should all be circles, no matter what the copied figures in the first plane might be. (Perhaps this is the most convenient form of representation for us.) That is, different figures on I are mapped onto II by different methods of projection. In order in this case to construe the circles in II as representations of the figures in I, I shall have to give the method of projection for each circle; the mere fact that a figure in I is represented as a circle in II I by itself tells us nothing about the shape of the figure copied. That an image in II is a circle is just the established norm of our mapping. - Well, the same thing happens when we depict reality in our language in accordance with the subject-predicate form. The subject-predicate form serves as a projection of countless different logical forms. Frege's "Concept and Object" is the same as subject and predicate. If a table is painted brown, then it's easy to think of the wood as bearer of the property brown and you can imagine what remains the same when the colour changes. Even in the case of one particular circle which appears now red, now blue. It is thus easy to imagine what is red, but difficult to imagine what is circular. What remains in this case if form and colour alter? For position is part of the form and it is arbitrary for me to lay down that the centre should stay fixed and the only changes in form be changes in the radius. We must once more adhere to ordinary language and say that a patch is circular. It is clear that here the phrase "bearer of a property" in this context conveys a completely wrong - an impossible - picture. IfI have a lump of clay, I can consider it as the bearer of a form, and that, roughly, is where this picture comes from. 1. I have here corrected an inadvertent transposition of "I" and "II" in Wittgenstein's German. (Trs.)

204

"The patch changes its form" and "the lump of clay changes its form" are different forms of propositions. You can say "Measure whether that is a circle" or "See whether that over there is a hat". You can also say "Measure whether that is a circle or an ellipse", but not" ... whether that is a circle or a hat"; nor "See whether that is a hat or red". IfI point to a curve and say "That is a circle" then someone can object that if it were not a circle it would no longer be that. That is to say, what I mean by the word "that" must be independent of what I assert about it. ("Was that thunder, or gunfire?" Here you could not ask "Was that a noise ?") How are two circles of the same size distinguished? This question makes it sound as if they were pretty nearly one circle and only distinguished by a nicety. In the technique of representation by equations what is common is expressed by the form of the equation, and the difference by the difference in the coordinates of the centres. So it is as if what corresponds with the objects falling under the concept were here the coordinates of the centres. Couldn't you then say, instead of"This is a circle", "This point is the centre of a circle"? For to be the centre of a circle is an external property of the point. What is necessary to a description that - say - a book is in a certain position? The internal description of the book, i.e. of the concept, and a description of its place which it would be possible to give by giving the co-ordinates of three points. The proposition "Such a book is here" would mean that it had these three co-ordinates. For the specification of the "here" must not prejudge what is here. But doesn't it come to the same thing whether I say "This is a book" or "Here is a book" ? The proposition would then amount

206

to saying, "These are three corners of such a book". Similarly you can also say "This circle is the projection of a sphere" or "This is a man's appearance". All that I am saying comes to this, that (x) must be an external description of x. If in this sense I now say in three-dimensional space "Here is a circle" and on another occasion "Here is a sphere" are the two "here's" of the same type? I want to ask: can one ~ignificantly say of the same 'object': it is a circle, and: it is a sphere? Is the subject of each of these predicates of the same type? Both could be the three coordinates of the relevant centre-point. But the position of the circle in three-dimensional space is not fixed by the coordinates of its centre. On the other hand you can of course say "It's not the noise, but the colour that makes me nervous" and here it might look as if a variable assumed a colour and a noise as values. ("Sounds and colours can be used as vehicles of communication".) It is clear that this proposition is of the same kind as "if you hear a shot, or see me wave, run". For this is the kind of co-ordination on the basis of which a heard or seen language functions. "Is it conceivable that two things have all their properties in common ?" - If it isn't conceivable, then neither is its opposite. We do indeed talk about a circle, its diameter, etc. etc., as if we were describing a concept in complete abstraction from the objects falling under it. - But in that case 'circle' is not a predicate in the original sense. And in general geometry is the place where concepts from the most different regions get mixed up together.

3 Oijects

"In a certain sense, an object cannot be described." (So too Plato: "You can't give an account of one but only name it.") Here "object" means "reference of a not further definable word", and "description" or "explanation" really means: "definition". For of course it isn't denied that the object can be "described from outside", that properties can be ascribed to it and so on. So when we use the proposition above we are thinking of a calculus with signs or names that are indefinable - or, more accurately, undefined - and we are saying that no account can be given of them. "What a word means a proposition cannot tell." What is the distinction, then, between blue and red? We aren't of the opinion that one colour has one property and the other another. In any case, the properties of blue and red are that this body (or place) is blue, and that other is red. When asked "what is the distinction between blue and red?" we feel like answering: one is blue and the other red. But of course that means nothing and in reality what we're thinking of is the distinction between the surfaces or places that have these colours. For otherwise the question makes no sense at all. Compare the different question: "What is the distinction between orange and pink ?" One is a mixture of yellow and red, the other a mixture of white and red. And we may say accordingly: blue comes from purple when it gets more bluish, and red comes from purple when that gets more and more reddish.

2.08

So what I am saying means: red can't be described. But can't we represent it in painting by painting something red? No, that isn't a representation in painting of the meaning of the word 'red' (there's no such thing). The portrait of red. Still, it's no accident that in order to define the meaning of the word "red" the natural thing is to point to a red object. (What is natural about it is portrayed in that sentence by the double occurrence of the word 'red'). To say that blue is on the bluish side of blue-red and red on the reddish side is a grammatical sentence and therefore akin to a definition. And indeed one can also say: more bluish = more like blue. "If you call the colour green an object, you must be saying that it is an object that occurs in the symbolism. Otherwise the sense of the symbolism, and thus its very existence as a symbolism, would not be guaranteed." But what does that assert about green, or the word "green"? ((That sentence is connected with a particular conception of the meaning-relation and a particular formulation of the problem the relation raises».

4

Elementary Propositions Al Can a logical product be hidden in a proposition? And if so, how does one tell, and what methods do we have of bringing the hidden element of a proposition to light? If we haven't yet got a method, then we can't speak of something being hidden or possibly hidden. And if we do have a method of discovery then the only way in which something like a logical product can be hidden in a proposition is the way in which a quotient like 753/3 is hidden until the division has been carried out. The question whether a logical product is hidden in a sentence is a mathematical problem. So an elementary proposition is a proposition which, in the calculus as I am now using it, is not represented as a truth-function of other sentences.

1,

!

The idea of constructing elementary proposltlons (as e.g. Carnap has tried to do) rests on a false notion of logical analysis. It is not the task of that analysis to discover a theory of elementary propositions, like discovering principles of mechanics. My notion in the Tractatus Logico-Philosophicus was wrong: I) because I wasn't clear about the sense of the words "a logical product is hidden in a sentence" (and suchlike), 2) because I too thought that logical analysis had to bring to light what was hidden (as chemical and physical analysis does). The proposition "this place is now red" (or "this circle is now red") can be called an elementary proposition if this means that it is

1.

From the I93z(?) typescript where it appears as a chapter by itself.

210

neither a truth-function of other propositions nor defined as such. (Here I am disregarding combinations such as p. : qv ~ q and the like.) But from "a is now red" there follows "a is now not green" and so elementary propositions in this sense aren't independent of each other like the elementary propositions in the calculus I once described - a calculus to which, misled as I was by a false notion of reduction, I thought that the whole use of propositions must be reducible.

If you want to use the appellation "elementary proposition" as I did in the Tractatus Logico-Philosophicus and as Russell used "atomic proposition", you may call the sentence "Here there is a red rose" an elementary proposition. That is to say, it doesn't contain a truth-function and it isn't defined by an expression which contains one. But if we're to say that a proposition isn't an elementary proposition unless its complete logical analysis shows that it isn't built out of other propositions by truth-functions, we are presupposing that we have an idea of what such an 'analysis' would be. Formerly, I myself spoke of a 'complete analysis', and I used to believe that philosophy had to give a definitive dissection of propositions so as to set out clearly all their connections and remove all possibilities of misunderstanding. I spoke as if there was a calculus in which such a dissection would be possible. I vaguely had in mind something like the definition that Russell had given for the definite article, and I used to think that in a similar way one would be able to use visual impressions etc. to define the concept say of a sphere, and thus exhibit once for all the connections between the concepts and lay bare the source of all misunderstandings, etc. At the root of all this there was a false and idealized picture of the use of language. Of course, in particular cases one can J

I. From a later MS note book, probably written in summer I936, some two years after the main text of this volume. 211

clarify by definitions the connections between the different types of use of expressions. Such a definition may be useful in the case of the connection between 'visual impression' and 'sphere'. But for this purpose it is not a definition of the concept of a physical sphere that we need; instead we must describe a language game related to our own, or rather a whole series of related language games, and it will be in these that such definitions may occur. Such a contrast destroys grammatical prejudices and makes it possible for us to see the use of a word as it really is, instead of inventing the use for the word. There could perhaps be a calculus for dissecting propositions; it isn't hard to imagine one. Then it becomes a problem of calculation to discover whether a proposition is or is not an elementary proposition. The question whether e.g. a logical product is hidden in a sentence is a mathematical problem. - What "hidden" means here is defined by the method of discovery (or, as it might be, by the lack of a method).

What gives us the idea that there is a kind of agreement between thought and reality? - Instead of "agreement" here one might say with a clear conscience "pictorial character". I But is this pictorial character an agreement? In the Tractatus Logico-Philosophicus I said something like: it is an agreement of form. But that is an error. First of all, "picture" here is ambiguous. One wants to say that an order is the picture of the action which was carried out on the order; but also, a picture of the action which is to be carried out as an order.

We may say: a blueprint serves as a picture of the object which the workman is to make from it. And here we might call the way in which the workman turns such a drawing into an artefact "the method of projection". We might now express ourselves thus: the method of projection mediates between the drawing and the object, it reaches from the drawing to the artefact. Here we are comparing the method of projection with projection lines which go from one figure to another. - But if the method of projection is a bridge, it is a bridge which isn't built until the application is made. - This comparison conceals the fact that the picture plus the projection lines leaves open various methods of application; it makes it look as if what is depicted, even if it does not exist in fact, is determined by the picture and the projection lines in an ethereal manner; every bit as determined, that is to say, as if it did exist. (It is 'determined give or take a yes or no.') In that case what we may call 'picture' is the blueprint plus the method of its application. And we now imagine the method as something which is attached to the blueprint whether or not it is used. (One can "describe" an application even if it doesn't exist). Now I would like to ask "How can the blueprint be used as a representation, unless there is already an agreement with what is to be made?" - But what does that mean? Well, perhaps this: how could I play the notes in the score on the piano if they didn't already have a relationship to particular types of movement of the hand? Of course such a relationship sometimes consists in a certain agreement, but sometimes not in any agreement, but merely in our having learnt to apply the signs in a particular way. What the comparison between the method of projection and the projection lines connecting the picture with the object does is to make all these cases alike - because that is what attracts us. You may say: I count the projection lines as part of the picture - but not the method of projection.

I.

212

Cf. Tractatus

2.

151 3 (Editor).

You may of course also say: I count a description of a method of projection as part of the picture. So I am imagining that the difference between proposition and reality is ironed out by the lines of projection belonging to the picture, the thought, and that no further room is left for a method of application, but only for agreement and disagreement.

Is time essential to propositions? Comparison between time and truthjunctions

If we had grammar set out in the form of a book, it wouldn't be a series of chapters side by side, it would have quite a different structure. And it is here, if I am right, that we would have to see the distinction between phenomenological and non-phenomenological. There would be, say, a chapter about colours, setting out the rules for the use of colour-words; but there would be nothing comparable in what the grammar had to say about the words "not", "or", etc. (the "logical constants"). It would, for instance, be a consequence of the rules, that these latter words unlike the colour words were usable in every proposition; and the generality belonging to this "every" would not be the kind that is discovered by experience, but the generality of a supreme rule of the game admitting of no appeal. How does the temporal character of facts manifest itself? How does it express itself, if not by certain expressions having to occur in our sentences? That means: how does the temporal character of facts express itself, if not grammatically? "Temporal character" - that doesn't mean that I come at 5 o'clock, but that I come at some time or other, i.e. that my proposition has the structure it has. We are inclined to say that negation and disjunction are connected with the nature of the proposition, but that time is connected with its content rather than with its nature. But if two things are equally universal, how can it show itself in grammar that one of them is connected with the nature of the proposition and the other is not? Or should I have said that time is not equally universal since mathematical propositions can be negated and occur in disjunctions, without being temporal? There is indeed a connection here, though this form of portraying the matter is misleading.

214

But that shows what I mean by "proposition." or "nature of the proposition" . Why - I want to ask - is the temporal character of propositions so universal? Might one also put the question thus: "How does it happen that every fact of experience can be brought into a relationship with what is shown by a clock ?" Having two kinds of generality in the way I spoke of would be as strange as if there were two equally exceptionless rules of a game and one of them were pronounced to be more fundamental. As if one could ask whether in chess the king or the chess board was more important; which of the two was more essential, and which more accidental. There's at least one question that seems in order: suppose I had written up the grammar, and the different chapters on the colour words, etc. etc. were there one after the other, like rules for each of the chess pieces, how would I know that those were all the chapters? If there turns out to be a common property in all the chapters so far in existence, we seem to have encountered a logical generality that is not an essential, i.e. a priori generality. But we can't say that the fact that chess is played with 16 pieces is any less essential to it than its being played on a chessboard. Since time and the truth-functions taste so different, and since they manifest their nature only and wholly in grammar, it is grammar that must explain the different taste. . One tastes like content, the other like form of representatlOn. They taste as different as a plan and a line through a plan.

216

It appears to me that the present, as it occurs in the proposition "the sky is blue" (if this proposition isn't meant as a hypothesis), is not a form of time, so that the present in this sense is atemporal.

Does time enter into a landscape picture? or into a still life ? Literature consisting of descriptions of landscapes. It is noteworthy that the time of which I am here speaking is not time in a physical sense. We are not concerned with measuring time. It is fishy that something which is unconnected with measurement is supposed to have a role in propositions like that of physical time in the hypotheses of physics. Discuss: The distinction between the logic of the content and the logic of the propositional form in general. The former seems, so to speak, brightly coloured, and the latter plain; the former seems to be concerned with what the picture represents, the latter to be a characteristic of the pictorial form like a frame. By comparison with the way in which the truth-functions are applicable to all propositions,. it seems to us accidental that all propositions contain time in some way or other. The former seems to be connected with their nature as propositions, the latter with the nature of the reality we encounter. ((Added later in the margins» A sentence can contain time in very different senses. You are hurting me. The weather is marvellous outside. The Inn flows into the Danube. Water freezes at 0°. I often make slips of the pen Some time ago ... I hope he will come. At 5 o'clock.

21 7

This kind of steel is excellent. The earth was once a ball of gas.

6

The Nature of Hypotheses

You could obviously explain an hypothesis by means of pictures. I mean, you could e.g. explain the hypothesis "there is a book lying here" with pictures showing the bookin plan, elevation and various cross-sections. Such a representation gives a law. Just as the equation of a curve gives a law, by means of which you may discover the ordinates, if you cut at different abscissae. In which case the verifications of particular cases correspond to cuts that have actually been made. If our experiences yield points lying on a straight line, the proposition that these experiences are various views of a straight line is an hypothesis. The hypothesis is a way of representing this reality, for a new experience may tally with it or not, or possibly make it necessary to modify the hypothesis. If for instance we use a system of coordinates and the equation for a sphere to express the proposition that a sphere is located at a certain distance from our eyes, this description has a greater multiplicity than that of a verification by eye. The first multiplicity corresponds not to one verification but to a law obeyed by verifications. An hypothesis is a law for forming propositions. You could also say: an hypothesis is a law for forming expectations. A proposition is, so to speak, a particular cross-section of an hypothesis. According to my principle two suppositions must have the same sense if every possible experience that confirms the one also con-

Zl8

21 9

firms the other, if, that is, no decision between the two is conceivable on the basis of experience.

propositions aren't true that were implicit in the first. But it isn't as if the first turns out to have been a logical product all along.

The representation of a curve as a straight line with deviations. The equation of the curve includes a parameter whose course expresses the deviations from a straight line. It isn't essential that these deviations should be "slight". They can be so large that the curve doesn't look like a straight line at all. "Straight line with deviations" is only one form of description. It makes it easier for me to eliminate, or neglect, a particular component of the description if I so wish. (The form "rule with exceptions").

The best comparison for every hypothesis, - something that is itself an example of an hypothesis - is a body in relation to a systematic series of views of it from different angles.

What does it mean, to be certain that one has toothache? (If one can't be certain, then grammar doesn't allow the use of the word "certain" in this connection.) The grammar of the expression "to be certain". We say "If I say that I see a chair there, I am saying more than I know for certain". And commonly that means "But all the same, there's one thing that I do know for certain." But if we now try to say what it is, we find ourselves in a certain embarrassment. "I see something brown - that is certain." That's meant to say that the brown colour is seen and not perhaps merely conjectured from other symptoms. And we do indeed say quite simply: "I see something brown." If someone tells me "Look into this telescope, and make me a sketch of what you see", the sketch I make is the expression of a proposition, not of a hypothesis.

If I say "Here there is a chair", I mean more - people say - than the mere description of what I perceive. This can only mean that that proposition doesn't have to be true, even though the description fits what is seen. Well, in what circumstances would I say that that proposition wasn't true? Apparently, if certain other

220

Making a discovery in a scientific investigation (say in experimental physics) is of course not the same thing as making a discovery in ordinary life outside the laboratory; but the two are similar and a comparison with the former can throw light on the latter. There is an essential distinction between propOSltlOnS like "That is a lion", "The sun is larger than the earth", and propositions like "Men have two hands". Propositions like the first pair contain a "this", "now", "here" and thus connect immediately with reality. But if there happened to be no men around, how would I go about checking the third proposition? It is always single faces of hypotheses that are verified. Perhaps this is how it is: what an hypothesis explains is itself only expressible by an hypothesis. Of course, this amounts to asking whether there are any primary propositions that are definitively verifiable and not merely facets of an hypothesis. (That is rather like asking: are there surfaces that aren't surfaces of bodies?) At all events, there can't be any distinction between an hypothesis used as an expression of an immediate experience and a proposition in the stricter sense. There is a distinction between a proposition like "Here there is a sphere in front of me" and "It looks as if there is a sphere in front of me". The same thing shows itself also thus: one can say "There seems to be a sphere in front of me", but it is senseless to say "It looks as if there seems to be a sphere here". So too one can say

22I

~

t

,

I

"Here there is probably a sphere", but not "Here there probably appears to be a sphere". In such a case people would say" After all, you must know whether there appears to be". There is nothing hypothetical in what connects the proposition with the given faci:o It's clear that reality - I mean immediate experience - will sometimes give an hypothesis the answer yes, and sometimes the answer no (here of course the "yes" and "no" express only confirmation and lack of confirmation); and it's clear that these affirmations and denials can be given expression. The hypothesis, if that face of it is laid against reality, becomes a proposition. It may be doubtful whether the body I see is a sphere, but it can't be doubtful that from here it looks to be something like a sphere. - The mechanism of hypothesis would not function if appearance too were doubtful so that one couldn't verify beyond doubt even a facet of the hypothesis. If there were a doubt here, what could take the doubt away? If this connection too were loose, there would be no such thing as confirming an hypothesis and it would hang entirely in the air, quite pointless (and therefore senseless). If I say "I saw a chair", that (in one sense) isn't contradicted by the proposition "there wasn't one there". For I could use the first proposition in the description of a dream and then nobody would use the second to contradict me. But the description of the dream throws a light on the sense of the words "I saw". Again, in the proposition "there wasn't one there", the word "there" may have more than one meaning.

222

I am in agreement with the opinions of contemporary physicists when they say that the signs in their equations no longer have any "meanings" and that physics cannot attain to any such meanings, but must stay put at the signs. But they don't see that the signs have meaning in as much as - and only in as much as - observable phenomena do or do not correspond to them, in however circuitous a manner. Let us imagine that chess had been invented not as a board game, but as a game to be played with numbers and letters on paper, so that no one had ever imagined a board with 64 squares in connection with it. And now suppose someone made the discovery that the game corresponded exactly to a game which could be played on a board in such and such a way. This discovery would have been a great simplification of the game (people who would earlier have found it too difficult could now play it). But it is clear that this new illustration of the rules of the game would be nothing more than a new, more easily surveyable symbolism, which in other respects would be on the same level as the written game. Compare with this the talk about physics nowadays not working with mechanical models but "only with symbols".

7

Probabiliry

The probability of an hypothesis has its measure in how much evidence is needed to make it profitable to throw it out. It's only in this sense that we can say that repeated uniform experience in the past renders the continuation of this uniformity in the future probable. If, in this sense, I now say: I assume the sun will rise again tomorrow, because the opposite is so unlikely, I here mean by "likely" and "unlikely" something completely different from what I mean by these words in the proposition "It's equally likely that I'll throw heads or tails". The two meanings of the word "likely" are, to be sure, connected in certain ways, but they aren't identical.

A postulate must be such that no conceivable experience can refute it, even though it may be extremely inconvenient to cling to the postulate. To the extent to which we can talk here of greater or slighter convenience, there is a greater or slighter probability of the postulate.

We only give up an hypothesis for an ever higher gain. Induction is a process based on a principle of economy. The question how simple a representation is yielded by assuming a particular hypothesis is directly connected, I believe, with the question of probability. We may compare a part of an hypothesis with the movement of a part of a gear, a movement that can be stipulated without prejudicing the intended motion. But then of course you have to make appropriate adjustments to the rest of the gear if it is to produce the desired motion. I'm thinking of a differential gear. - Once I've decided that there is to be no deviation from a certain part of my hypothesis no matter what the experience to be described may be, I have stipulated a mode of representation and this part of my hypothesis is now a postulate.

It's senseless to talk of a measure for this probability at this juncture. The situation here is like that in the case of two kinds of numbers where we can with a certain justice say that the one is more like the other (is closer to it) than a third, but there isn't any numerical measure of the similarity. Of course you could imagine a measure being constructed in such cases, too, say by counting the postulates or axioms common to the two systems, etc. etc. I give someone the following piece of information, and no more: at such and such a time you will see a point of light appear in the interval AB. B c A

I

I

I

Does the question now makes sense "Is it more likely that this point will appear in the interval AC than in CB"? I believe, obviously not. - I can of course decide that the probability of the event's happening in CB is to be in the ratio CBjAC to the proba225

bility of its happening in AC; however, that's a decision I can have empirical grounds for making, but about which there is nothing to be said a priori. It is possible for the observed distribution of events not to lead to this assumption. The probability, where infinitely many possibilities come into consideration, must of course be treated as a limit. That is, if I divide the stretch AB into arbitrarily many parts of arbitrary lengths and regard it as equally likely that the event should occur in anyone of these parts, we immediately have the simple case of dice before us. And now I can - arbitrarily - lay down a law for constructing parts of equal likelihood. For instance, the law that, if the lengths of the parts are equal, they are equally likely. But any other law is just as permissible. Couldn't I, in the case of dice too, take, say, five faces together as one possibility, and oppose them to the sixth as the second possibility? And what, apart from experience, is there to prevent me from regarding these two possibilities as equally likely? Let's imagine throwing, say, a red ball with just one very small green patch on it. Isn't it much more likely in this case for the red area to strike the ground than for the green? - But how would we support this proposition? Presumably by showing that when we throw the ball, the red strikes the ground much more often than the green. But that's got nothing to do with logic. - We may always project the red and green surfaces and what befalls them onto a surface in such a way that the projection of the green surface is greater than or equal to the red; so that the events, as seen in this projection, appear to have a quite different probability ratio from the one they had on the original surface. If, e.g. I reflect the events

in a suitably curved mirror and now imagine what I would have held to be the more probable event if! had only seen the image in the mirror. The one thing the mirror can't alter is the number of clearly demarcated possibilities. So that if I have n coloured patches on my ball, the mirror will also show n, and if I have decided that these are to be regarded as equally likely, then I can stick to this decision for the mirror image too. To make myself even clearer: if I carry out the experiment with a concave mirror, i.e. make the observations in a concave mirror, it will perhaps then look as if the ball falls more often on the small surface than on the much larger one; and it's clear that neither experiment - in the mirror or outside it - has a claim to precedence. We may apply our old principle to propositions expressing a probability and say, we shall discover their sense by considering what verifies them. If I say "That will probably occur", is this proposition verified by the occurrence or falsified by its non-occurrence? In my opinion, obviously not. In that case it doesn't say anything about either. For if a dispute were to arise as to whether it is probable or not, it would always be arguments from the past that would be adduced. And this would be so even when what actually happened was already known. Causality depends on an observed uniformity. This does not mean that a uniformity so far observed will always continue, but what cannot be altered is that the events so far have been uniform; that can't be the uncertain result of an empirical series which in its turn isn't something given but something dependent on another uncertain one and so on ad infinitum. When people say that the proposition "it is probable that p will occur" says something about the event p, they forget that the probability remains even when the event p does not occur.

226

227

The proposltlon "p will probably occur" does indeed say something about the future, but not something "about the event p", as the grammatical form of the statement makes us believe.

A Galtonian photograph is the picture of a probability. The law of probability is the natural law you see when you screw up your eyes.

If! ask for the grounds of an assertion, the answer to the question holds not only for this person and for this action (assertion), but quitegeneral&.

"On average, the points yielded by the experiment lie on a straight line". "If I throw with a good die, then on average I throw a one every six throws". What does that mean? Is the proposition compatible with a'!Y experience I may have? If so, it says nothing. Have I decided in advance which experiences are incompatible with it and what is the limit beyond which exceptions may not go without upsetting the rule? No. But couldn't I have set such a limit? Of course. - Suppose that the limit had been set thus: if 4 out of 6 successive throws turn out the same, then it's a bad die. Now someone says: "But if that happens only very seldom, mayn't it be a good one after all?" - To that the answer is as follows. If I permit the turning up of 4 similar throws among 6 successive ones to occur within a certain number of throws, then I am replacing the first limit with a different one. But if I say "any number of similar successive throws is allowed, as long as it happens sufficiently rarely", then strictly speaking I've defined the goodness of the die in a way that makes it independent of the result of the throws; unless by the goodness of a die I do not mean a property of the die, but a property of a particular game played with it. In that case I can certainly say: in any game I call the die good provided that among the N throws of the game there occur not more than log N similar successive throws. However, that doesn't give a test for the checking of dice, but a criterion for judging a particular game.

If I say "the weather looks like rain" do I say anything about future weather? No; I say something about the present weather, by means of a law connecting weather at any given time with weather at an earlier time. This law must already be in existence, and we are using it to construct certain statements about our experience. We might say the same of historical statements too. But I was too quick to say that the proposition "the weather looks like rain" says nothing about future weather. It all depends what is meant by "saying something about something". The sentence says just what it says. The sentence "p will probably occur" says something about the future only in a sense in which its truth and falsehood are completely independent of what will happen in the future. If we say: "the gun is now aiming at the point p" we aren't saying anything about where the shot will hit. Giving the point at which it is aiming is a geometrical means of assigning its direction. That this is the means we use is certainly connected with certain observations (projectile parabolas, etc.) but these observations don't enter into our present description of the direction. Parabola

228

We say that if the die is quite regular and isn't interfered with then the distribution of the numbers I, 2, 3, 4, 5, 6 among the throws must be uniform, since there is no reason why one number should occur more often than another.

But now let's represent the throws by the values of the function (x - 3)2 for the arguments I to 6, i.e. by the numbers 0, I, 4, 9 instead of by the numbers I to 6. Is there a reason why one of these numbers should turn up in the new results more often than another? This shows us that the a priori law of probability, like the minimum-principles of mechanics etc., is a form that laws may take. If it had been discovered by experiment that the distribution of the throws I to 6 with a regular die was such that the distribution of the values of (x - 3)2 was uniform, it would have been this regularity that was defined as the a priori regularity. We do the same thing in the kinetic theory of gases: we represent the distribution of molecular movements in the form of some sort of uniform distribution; but we make the choice of what is uniformly distributed - and in the other case of what is reduced to a minimum - in such a way that our theory agrees with experience.

I may say that a stretch gives the general impression of a straight line; but I cannot say: "This bit of line looks straight, for it could be a bit of a line that as a whole gives me the impression of being straight." (Mountains on the earth and moon. The earth a ball.) An experiment with dice lasts a certain time, and our expectations about future throws can only be based on tendencies we observe in what happens during this experiment. That is to say, the experiment can only give grounds for expecting that things will go in in the way shown by the experiment; but we can't expect that the experiment, if continued, will now yield results that tally better with a preconceived idea of its course than did those of the experiment we have actually performed. So if, for instance, I toss a coin and find no tendency in the results of the experiment itself for the number of heads and tails to approximate to each other more closely, then the experiment gives me no reason to suppose that if it were continued such an approximation would emerge. Indeed, the expectation of such an approximation must itself refer to a definite point in time, since we can't say we're expecting something to happen eventuallY, in the infinite future.

"The molecules move purely according to the laws of probability" is supposed to mean: physics gets out of the way, and now the molecules move as it were purely according to laws of logic. This idea is similar to the idea that the law of inertia is an a priori proposition: there too one speaks of what a body does when it isn't interfered with. But what is the criterion for its not being interfered with? Is it ultimately that it moves uniformly in a straight line? Or is it something different? If the latter, then it's a matter of experience whether the law of inertia holds; if the former, then it wasn't a law at all but a definition. So too with the proposition, "if the particles aren't interfered with, then the distribution of their motions is such and such". What is the criterion for their not being interfered with? etc.

Any "reasonable expectation" is an expectation that a rule we have observed up to now will continue to hold. (But the rule must have been observed and can't, for its part too, be merely expected.)

To say that the points yielded in this experiment lie roughly on this line, e.g. a straight line, means something like: "seen for this distance, they seem to lie on a straight line".

A ray is emitted from the light source S striking the surface AB to form a point of light there, and then striking the surface AB'. We have no reason to suppose that the point on AB lies to the left or to the right of M, and equally none for supposing that the

The logic of probability is only concerned with the state of expectation in the sense in which logic in general is concerned with thinking.

".,.."""

....- - -

--

e::;::-- -::. - - - -

S .........

-

-

- -

.... "-

.....

A point on AB' lies on one side or the other of m. This yi~lds therefore incompatible probabilities. But if I make an assu:npt~on about ~he probability of the point on AB lying in AM, how ~s thlS assumpt~on verified? Surely, we think, by a frequency experiment. Supposmg this confirms the view that the probabilities of AM and BM are equal (and so the probabilities of Am and B'n: differ), then it is recognized as the right one and thus shows itself to be an hypothesis belonging to physics. The geometrical construction merely shows that the fact that AM = MB was no ground for assuming equal likelihood. Suppose that measurement shows the die to be accurate and regular, that the numbers on its sides don't influence the throws, and that it is thrown by a hand whose movements follow no definite rules: does it follow that the distribution among the throws of each of the throws from I to 6 will be uniform on average? Where is the uniform distribu:ion s,uppose~ to come from? The accuracy and regularity of the dle can t estabhsh that t~e distribution of throws will be uniform on average. (It would be, as lt were, a monochrome premise with a mottle conclusion.) And we

haven't made any SUpposltlOns about the movements while throwing. (Making the bundles of hay equal gives reason to believe that the donkey will starve to death between them; it doesn't give reason to believe that he will eat from each with roughly equal frequency.) - It is perfectly compatible with our assumptions for one hundred ones to be thrown in succession, if friction, hand-movements and air-resistance coincide appropriately. The experimental fact that this never happens is a fact about those factors, and the hypothesis that the throws will be uniformly distributed is an hypothesis about the operation of those factors . Suppose someone says that a lever with arms of equal length must remain at rest under the influence of equal and opposite forces, since there is no cause to make it move to one side rather than to the other. That only means that if the lever moves to one side after we have ascertained the equality of the arms and the equal and opposite nature of the forces, then we can't explain this on the basis of the preconditions we know or have assumed. (The form that we call "explanation" must be asymmetrical: like the operation which makes" 2a + 3b" out of "a + b"). But on the basis of our presuppositions we can indeed explain the lever's continuance at rest. - Could we also explain a swing to left and right with roughly equal frequency? No, because once again the swing involves asymmetry; we would only explain the symmetry in this asymmetry. If the lever had rotated to the right with a uniform motion, one could similarly have said: given the symmetry of the conditions I can explain the uniformity of the motion, but not its direction. A lack of uniformity in the distribution of the throws is not to be explained by the symmetry of the die. It is only to this extent that the symmetry explains the uniformity of the distribution. - For one can of course say: if the numbers on the sides of the die have no effect, then the difference between them cannot explain an irregularity in the distribution; and of course similar circumstances can't explain differences; and so to that extent one might infer a regularity. But in that case why is there any difference at

all between different throws? Whatever explains that must also explain their approximate regularity. It's just that the regularity of the die doesn't interfere with that regularity.

frequency actually observed. And that, of course, is an absolute frequency.

Suppose that a man throwing dice every day threw nothing but ones for a week, using dice that proved good by every other method of testing and that gave the usual results when thrown by others. Has he grounds, now, for supposing that there is a law of nature that he will always throw ones? Has he grounds for believing that it will go on like this, or has he grounds for believing that this regularity can't last much longer? Has he reason to abandon the game since it has become clear that he can only throw ones, or reason to play on since in these circumstances it is all the more probable that he will throw a higher number at the next throw? In actual fact, he will refuse to accept the regularity as a natural law: at least, it will have to go on for a long time before he will entertain the possibility. But why? I believe it is because so much of his previous experience in life speaks against there being a law of nature of such a sort, and we have - so to speak - to surmount all that experience, before embracing a totally new way of looking at things. If we infer from the relative frequency of an event its relative frequency in the future, we can of course only do that from the frequency which has in fact been so far observed. And not from one we have derived from observation by some process or other for calculating probabilities. For the probability we calculate is compatible with a'!J frequency whatever that we actually observe, since it leaves the time open. When a gambler or insurance company is guided by probability, they aren't guided by the probability calculus, since one can't be guided by this on its own, because a'!Jthing that happens can be reconciled with it: no, the insurance company is guided by a

235

8

The concept "about" Problem of the "heap" "He came from about there --+." "About there is the brightest point of the horizon". "Make the plank about 2 m long". In order to say this, must I know of limits which determine the margin of tolerance of this length? Obviously not. Isn't it enough e.g. to say "A margin of ±I cm is perfectly permissible; 2 would be too much"? - Indeed it's an essential part of the sense of my proposition that I'm not in a position to give "precise" bounds to the margin. Isn't that obviously because the space in which I am working here doesn't have the same metric as the Euclidean one? Suppose one wanted to fix the margin of tolerance exactly by experiment, by altering the length, approaching the limits of the margin and asking in each case whether such a length would do or not. After a few shortenings one would get contradictory results: at one time a point would be described as being within the limits, and at another time a point closer in would be described as impermissible, each time perhaps with the remark that the answers were no longer quite certain. It is the same sort of uncertainty as occurs in giving the highest point of a curve. We just aren't in Euclidean space and here there isn't a highest point in the Euclidean sense. The answer will mean "The highest point is about there" and the grammar of the word "about" - in this context - is part of the geometry of our space.

Surely it is like the way the butcher weighs things only to the nearest ounce, though that is arbitrary and depends on what are the customary counterweights. Here it is enough to know: it doesn't weigh more than PI and it doesn't weigh less than P 2· One might say: in principle giving the weight thus isn't giving a

number, but an interval, and the intervals make up a discontinuous series. Yet one might say: "at all events keep within ± I cm", thus setting an arbitrary limit. - If someone now said "Right, but that isn't the real limit of the permissible tolerance; so what is?" the answer would be e.g. "I don't know of any; I only know that ±2 is too much". Imagine the following psychological experiment.

-f(t::, The subject is shown curves g 1 g2 with a straight line A drawn across them. I will call the section of this line between gl and g2 a. Parallel to a we now draw b at an arbitrary distance and ask the subject whether he sees the section b as bigger than a, or cannot any longer distinguish between the two lengths. He replies that b seems bigger than a. Next we move closer to a, measuring half the distance from a to b and drawing c. "Do you see c as bigger than a ?" "Yes." - We halve the distance c-a and draw d. "Do you see d as bigger than a ?" "Yes." We halve a-d. "Do you see e as bigger than a?" - "No." - So we halve e-d. "Do you see f as bigger than e?" - "Yes." - So we halve e-f and draw h. We might approach the line a from the left hand side as well and then say that what corresponds in Euclidean space to a seen length a is not a single length but an interval of lengths, and in a similar way what corresponds to a single seen position of a line (say the pointer of an instrument) is an interval of positions in Euclidean space; but this interval has no precise limits. That means: it is bounded not by

237

points, but by converging intervals which do not converge upon a point. (Like the series of binary fractions that we get by throwing heads and tails.) The special thing about two intervals which are bounded in this blurred way instead of by points is that in certain cases the answer to the question whether they overlap or are quite distinct is "undecided"; and the question whether they touch, whether they have an end-point in common, is always a senseless one since they don't have end-points at all. But one might say "they have de facto end-points", in the sense in which the development of 1t has a de facto end. There is of course nothing mysterious about this property of "blurred" intervals; the somewhat paradoxical character is explained by the double use of the word "interval". The case is the same as that of the double use of the word "chess" to mean at one time the totality of the currently valid chess rules, and at another time the game invented in Persia by N. N. which developed in such and such a way. In one case it is nonsensical to talk of a development of the rules of chess and in another not. What we mean by "the length of a measured section" may be either what results from a particular measurement which I carry out today at 5 o'clock - in that case there is no " ± etc." for this assignment of length - or, something to which measurements approximate, etc. ; in the two cases the word "length" is used with quite different grammars. So too the word "interval" if what I mean by an interval is at one time something fixed and at another time something in flux. But we must not be surprised that an interval should have such a strange property; for we're now just using the word "interval" in a sense different from the usual one. And we can't say that we have discovered new properties of certain intervals, any more than we would discover new properties of the king in chess if we altered the rules of the game while keeping the designation "chess" and "king". (On the other hand cf. Brouwer on the law of excluded middle.)

I

rP\\~ III

~ VI

II

~ IV

~

V

~

VII

~ ~ I) II) III) IV) V) VI) VII)

the intervals are separate they are separate with a de facto contact undecided undecided undecided they overlap they overlap

So basically that experiment gives what we have called a "blurred" interval; on the other hand of course we could conceive experiments which would give a sharp interval instead. Suppose we moved a straight-edge from the starting position b, in the direction of a, keeping it parallel to b, until our subject began to display a particular reaction; in that case we could call the point at which the reaction first occurs the limit of our strip. Likewise we might of course call the result of a weighing "the weight of a ~ody" and in that sense there would be an absolutely exact weighIng, that is, one whose result did not have the forml"IW ± w".We would thus have altered the form of our expression, and we would have to say that the weight of bodies varied according to a law that was unknown to us. (The distinction between "absolutely exact" weighing and "essentially inexact weighing" is a grammatical distinction connected with two different meanings of the expression "result of weighing").

The indeterminacy of the word "heap". I could give as a definition: a body of a certain form and consistency etc. is a heap, if it has a volume of K cubic metres, or more; anything less than that I will call a heaplet. In that case there is no largest heaplet; that means, it is senseless to speak of a largest heaplet. Conversely, I could decide: whatever is bigger than K cubic metres is to be a heap, and in that case the expression "the smallest heap" has no meaning. But isn't this distinction an idle one? Certainly - if by the volume we mean a result of measurement in the normal sense; for such a result has the form "V ± v". But otherwise the distinction would be no more idle than the distinction between threescore apples and 61 apples. About the problem of the "heap": Here, as in similar cases, one might think that there is an official concept like the official length of a pace; say "A heap is anything that is bigger than half a cubic metre". But this would still not be the concept we normally use. For that there exists no delimitation (and if we fix one, we are altering the concept); it is just that there are cases that we count as within the extension of the concept, and cases that we no longer count as within the extension of the concept. "Make me a heap of sand here." - "Fine, that is certainly something he would call a heap." I was able to obey the command, so it was in order. But what about this command "Make me the smallest heap you would still call a heap"? I would say: that is nonsense; I can only determine a de facto upper and lower limit.

Part II On Logic and Mathematics

I LOGICAL INFERENCE

Is it because we understand the propositions that we know that q entails p ? Does a sense give rise to the entailment?

p.q. =.p means "q follows from p". q

(pvq)

1\ (p v q)" vq (p v q).q

p

q

pvq

q

T

T

T

T

T

T

T F

F

F

F

T

T

T T

T

T

T

F

F

F

F

F

F

(3x) . fx v fa. = .(3x) . fx, (3x) . fx. fa. = . fa. How do I know that? (Because for the equation above I gave a kind of proof). One might say something like: "I just understand '(3x).fx'''. (An excellent example of what "understand" means). But I might equally ask "How do I know that (3x) .fx follows from fa?" and answer "because I understand '(3x). fx'." But really how do I know that it follows? - Because that is the way I calculate. How do I know that (3x) . fx follows from fa ? Is it that I as it were see behind the sign "(3x) . fx", that I see the sense lying behind it and see from that that it follows from fa? Is that what understanding is? No, what that equation expresses is a part of the understanding (that is thus unpacked before my eyes). Compare the idea that understanding is first of all grasping in a flash something which then has to be unpacked like that. If I say "I know that (3x). fx follows, because I understand it" that would mean, that when I understand it, I see something different from the sign I'm given, a kind of definition of the sign which gives rise to the entailment. 2.43

Isn't it rather that the connection is set up and prescribed by th ';l " e equatlOns. For there IS no such thlng as a hidden connection. (3x).fx

fa

T

T

T

..,

F

..,.. ~

F

And why should the rule fE. fa. = . fa be an effect of another rule rather than being itself the primary rule? For what is "fE must somehow contain fa" supposed to mean? It doesn't contain it, in so far as we can work with fE without mentioning fa; but it does in so far as the rule fE. fa = . fa holds.

F

But, I used to think, mustn't (3x). fx be a truth function of fa for that to be possible, for that connection to be possible? Fo~ doe~n't (3x). Fx v Fa = (3x). fx simply say that fa is already

contalned In (3x). fx? Doesn't it show the connection between th~ fa and the (3x). fx? Not unless (3x). fx is defined as a logical sum (wlth fa as one of the terms of the sum). - If that is the case then (3x) . fx is merely an abbreviation. ' In logic there is no such thing as a hidden connection. You can't get behind the rules, because there isn't any behind. fE. fa. = fa. Can one say: that is only possible if fE follows from fa? Or must one say: that settles that fE is to follow from fa? If the for~er, it must be the structure that makes it follow, say because fE IS so defined as to have the appropriate structure. But can the entailment really be a kind of result of the visible structure of the signs, in the way that a physical reaction is the result of a ~hysical prop~rty? Doesn't it rather always depend on stipulations hke the equatlOn fE. fa. = . fa? Can it be read off from p v q that

244

it follows from p, or only from the rules Russell gives for the truthfunctions?

But the idea is that fE. fa. = fa can only hold in virtue of a definition of fE. That is, I think, because otherwise it looks, wrongly, as if a further stipulation had been made about fE after it had already been introduced into the language. But in fact there isn't any stipulation left for future experience to make. And the definition of fE in terms of "all particular cases" is no less impossible than the enumeration of all rules of the form fE.fx. = fx. Indeed the individual equations fE. fx. = fx are just precisely an expression of this impossibility. If we are asked: but is it now really certain that it isn't a different calculus being used, we can only say: if that means "don't we use other calculi too in our real language ?" I can only answer "I don't know any others at present". (Similarly, if someone asked "are these all the calculi of contemporary mathematics?" I might say "I don't remember any others, but I can read it up and find out more exactly"). But the question cannot mean "can no other calculus be used?" For how is the answer to that question to be discovered? A calculus exists when one describes it.

245

J

Can one say 'calculus' is not a mathematical concept? If I were to say "whether p follows from q must result from p and q alone": it would have to mean this: that p follows from q is a stipulation that determines the sense of p and q, not some extra truth that can be asserted about the sense of both of them. Hence one can indeed give rules of inference, but in doing so one is giving for the use of the written signs rules which determine their as yet undetermined sense; and that means simply that the rules must be laid down arbitrarily, i.e. are not to be read off from reality like a description. For when I say that the rules are arbitrary, I mean that they are not determined by reality in the way the description of reality is. And that means: it is nonsense to say that they agree with reality, e.g. that the rules for the words "blue" and "red" agree with the facts about those colours etc. What the equation p. q = p really shows is the connection between entailment and the truth-functions.

2

"Ifp follows from q, then thinking that q must involve thinking that p." Remember that a general proposition might entail a logical sum of a hundred or so terms, which we certainly didn't think of when we uttered the general proposition. Yet can't we say that it follows from it? "What follows from a thought must be involved in thinking it. For there is nothing in a thought that we aren't aware of while we are thinking it. It isn't a machine which might be explored with unexpected results, a machine which might achieve something that couldn't be read off from it. That is, the way it works is logical, it's quite different from the way a machine works. Qua thought, it contains nothing more than was put into it. As a machine functioning causally, it might be believed capable of anything; but in logic we get out of it only what we meant by it." If I say that the square is entirely white, I don't think of ten smaller rectangles contained in it which are white, and I can't think of "all" rectangles or patches contained in it. Similarly in the proposition "he is in the room" I don't think of a hundred possible positions he might be in and certainly not of all possible positions. "Wherever you hit the target you've won. You've hit it in the upper right hand section, so ... "

1.

Cf. Tractatus 5.

132

(Ed.).

At first sight there seem to be two kinds of deduction: in one of them the premise mentions everything the conclusion does and in the other not. An instance of the first kind is the inference from p. q to q; an instance of the second is the inference; the whole stick is white so the middle third of it is white too. This conclusion mentions bo~ndaries that are not mentioned in the first proposition. (That is dubious.) Again, if! say "If you hit the target anywhere in this circle you will win the prize ... " and then "You have hit it here, so ... " the place mentioned in the second proposition was not prescribed in the first. The target after the shot stands in a certain internal relation to the target as I saw it before, and that 247

relation consists in the shot's falling within the bounds of the general possibility that we foresaw. But the shot was not in itself foreseen and did not occur, or at least need not have occurred, in the first picture. For even supposing that at the time I thought of a thousand definite possibilities, it was at least possible for the one that was later realised to have been omitted. And if the foreseeing of that possibility really had been essential, the overlooking of this single case would have given the premise the wrong sense and the conclusion wouldn't any longer follow from it. On the other hand you don't add anything to the proposition "Wherever you hit this circle ... " by saying "Wherever you hit this circle, and in particular if you hit the black dot ... " If the black dot was already there when the first proposition was uttered, then of course it was meant too; and if it wasn't there, then the actual sense of the proposition has been altered by it.

If the criterion for p's following from q consists in "thinking of p being involved in thinking of q" then while thinking of the proposition "in this box there are 10 5 grains of sand", you are thinking also of the 10 5 sentences "In this box there is one grain of sand" " ... 2 grains of sand", etc. etc. What's the criterion here for the thought of one proposition's being involved in the thought of another? And what about a proposition like "There is a patch (F) between the limits AA" ?

1

1

1

B

A

•

F

1

1

1

A

B

C

Doesn't it follow from that that F is also between BB and CC and so on? Don't infinitely many propositions follow from a single one? Does that make it infinitely significant? - From the proposition "There is a patch between the limits AA" there follow as many propositions of the type "there is a patch between the limits BB" as I write out - and no more than I write out. Similarly, from p there follow as many propositions of the form p v ~ as I write out (or utter etc.). (A proof by induction proves as many propositions of the form ... as I write out.)

But what is it supposed to mean to say "If one proposition follows from another, thinking the second must involve thinking the first", since in the proposition "I am 170 cm tall" it isn't necessary to think of even a single one of the negative statements of height that follow from it? "The cross is situated thus on the straight line:

1

C

X 1"-

"So it is between the strokes".

"It is 16!0 here" - "So it is certainly more than 15 0" Incidentally, if you are surprised that one proposition can follow from another even though one doesn't think of the former while thinking of the latter, you should consider that p v q follows from p, and I certainly don't think all propositions of the form p v ~ while I am thinking p. The whole idea that a proposition has to be thought along with any proposition that entails it rests on a false, psychologising notion. We must concern ourselves only with what is contained in the signs and the rules.

249

L

3 The case of infinitefy matry propositions followingfrom a single one

I want to say that one proposition doesn't follow from another until it is confronted with it. The "etc ad infinitum" indicates only the possiblity of constructing propositions following from the first; it doesn't yield a definite number of such propositions. So mightn't I simply say: it is because it is impossible to write out infinitely many propositions (i.e. to say that is a piece of nonsense) that infinitely many propositions don't follow from a single proposition.

Is it impossible that infinitely many propositions should follow from a single one - in the sense, that is, that we might go on ad infinitum constructing new propositions from a single one according to a rule? Suppose that we wrote the first thousand propositions of the series in conjunction. Wouldn't the sense of this product necessarily approximate more closely to the sense of our first proposition than the product of the first hundred propositions? Wouldn't we obtain an ever closer approximation to the first proposition the further we extended the product? And wouldn't that show that it can't be the case that from one proposition infinitely many others follow, since I can't understand even the product with 10 10 terms and yet I understood the proposition to which the product with 10 100 terms is a closer approximation than the one with 10 10 terms?

B

A

I

I

1\

I

I

B'

What about the proposition "the surface is white from A to B"? It does follow from it that the surface is white from A'to B/. It needn't be a seen patch of white that is in question; and certainly the inference from the first proposition to the second is often drawn. Someone says to me "I have painted the patch white from A to B" and then I say "so it's certainly painted white from A'to B' ". It must be possible to say a priori that F(A'B') would follow from F(AB).

We imagine, perhaps, that the general proposition is an abbreviated expression of the product. But what is there in the product to abbreviate? It doesn't contain anything superfluous. If we need an example of infinitely many propositions following from a single one, perhaps the simplest is the way in which "a is red" entails the negation of all propositions that ascribe a different colour to a. The negative propositions are certainly not contained in the thought of the single positive one. Of course we might say that we don't distinguish infinitely many shades of colour; but the question is whether the number of shades of colour we distinguish has anything at all to do with the complexity of the first sentence: is it more or less complex the more or fewer colours we distinguish? Wouldn't this be what we'd have to say: it's only when a proposition exists that it follows from it. It's only when we have constructed ten propositions following from the first one that ten propositions do follow from it.

If the lines A' and B' exist, then the second proposition certainly does follow from the first (in that case the compositeness is already there in the first proposition); but in that case it is only as many propositions as correspond to its compositeness that follow from the first proposition (and so never infinitely many). "The whole is white, therefore a part bounded by such and such a line is white." "The whole was white, so that part of it also was white even if! didn't then perceive it bounded within it." "A surface seen as undivided has no parts." But let's imagine a ruler laid against the surface, so that the

251 I

~

appearance we are presented with is first I I and then I I I and then I I I I I. It doesn't at all follow from the first strip's being entirely white that in the second and the third everything except the graduating lines is white. "If you hit the target anywhere within the circle, you have won." "I think you will hit the target somewhere within the circle." Someone might ask about the first proposition: how do you know? Have you tried all possible places? And the answer would have to be: that isn't a proposition at all, it is a general stipulation.

What sort of proposition is: "On this strip you may see all shades of grey between black and white"? Here it looks at first glance as if we're talking about infinitely many shades. Indeed, we are apparently confronted here by the paradox that we can, of course, only distinguish a finite number of shades, and naturally the distinction between them isn't infinitely slight, and yet we see a continuous transition.

The inference doesn't go like this: "If the shot hits the target anywhere, you have won. You have hit the target there, so you have won". For where is this there? Is it marked out in any way other than by the shot - say by a circle? And was that already there on the target beforehand? If not, then the target has changed; if so, it must have been foreseen as a possible place to hit. We should rather say: "You have hit the target, so ... " The place on the target does not necessarily have to be given by a mark on the target, like a circle. For there are always descriptions like "nearer the centre", "nearer the edge", "on the right side at the top", etc. Wherever the target is hit such descriptions must always be possible. (But there are not "infinitely many" such descriptions. ) Does it make sense to say: "But if you hit the target, you must hit it somewhere" or "Wherever he hits the surface it won't be a surprise, we won't have to say 'I didn't expect that. I didn't know there was such a place'?" What that means is that it can't be a geometrical surprise.

It is just as impossible to conceive of a particular grey as being one of the infinitely many greys between black and white as it is to conceive of a tangent t as being one of the infinitely many transltional stages in going from t( to t 2. If I see a ruler roll around the circle from t( to t2 I see - if its motion is continuous - none of the intermediate positions in the sense in which I see t when the tangent is at rest; or else I see only a finite number of such positions. But if in such a case I appear to infer a particular case from a general proposition, then the general proposition is never derived from experience, and the proposition isn't a real proposition. If, e.g., I say "I saw the ruler move from t( to t2 therefore I must have seen it at t" this doen't give us a valid logical inference. That is, if what I mean is that the ruler must have appeared to me at t and so, if I'm talking about the position in visual space, then it doesn't

in the least follow from the premise. But if I'm talking about the physical ruler, then of course it's possible for the ruler to have skipped over position t and yet for the phenomenon in visual space to have remained continuous.

4

Can an experience show that one proposition follows from another?

The only essential point is that we cannot say that it was through experience we were made aware of an extra application of grammar. For in making that statement we would have to describe the application, and even if this is the first time I have realised that the description is true I must have been able to understand it even before the experience.

i.

Ii

I

It is the old question: how far can one now speak of an experience that one is not now having? What I cannot foresee I can not foresee. And what I can now speak of, I can now speak of independently of what I can't now speak of. Logic just is always complex. "How can I know everything that's going to follow?" What I can know then, I can also know now. But are there general rules of grammar, or only rules for general signs? What kind of thing in chess (or some other game) would count as a general rule or a particular rule? Every rule is general. Still, there is one kind of generality in the rule that p V q follows from p and a different kind in the rule that every proposition of the form p, ~ ~ p, ~ ~ ~ ~ p ... follows from p. q. But isn't the generality of the rule for the knight's move different from the generality of the rule for the beginning of a game? Is the word "rule" altogether ambiguous? So should we talk only about particular cases of rules, and stop talking about rules in general, and indeed about languages in general? "IfF I (a) [= ahas the colour F d entai1s~F2 (a) then the possibility of the second proposition must have been provided for in the

255 254

grammar of the first (otherwise how could we call FI and F2 colours ?) "If the second proposition as it were turned up without being expected by the first it couldn't possibly follow from it." "The first proposition must acknowledge the second as its consequence. Or rather they must be united in a single grammar which remains the same before and after the inference." (Here it is very difficult not to tell fairy tales about symbolic processes, just as elsewhere it is hard not to tell fairy tales about psychological processes. But everything is simple and familiar (there is nothing new to be discovered). That is the terrible thing about logic, that its extraordinary difficulty lies in the fact that nothing must be constructed, and everything is already present familiar.) "No proposition is a consequence of p unless p acknowledges it as its consequence." Whether a proposition entails another propOS1t10n must be clear from the grammar of the proposition and from that alone. It cannot be the result of any insight into a new sense: only of an insight into the old sense. It is not possible to construct a new proposition that follows from the old one which could not have been constructed (perhaps without knowing whether it was true or false) when the old one was constructed. If a new sense were discovered and followed from the first proposition, wouldn't that mean that that proposition had altered its sense?

II GENERALITY The proposition "The circle is in the square" is in a certain sense independent of the assignment of a particular position. (In a certain sense it is totallY unconnected.)

I would like to say: a general picture like same metric as a particular one.

I 0 I does not have the

In the general sign "101" the distances play no greater part than they do in the sign "aRb". The drawing I 0 I can be looked on as a representation of the "general case". It is as if it were not in a measurable space: the distances between the circle and the lines are of no consequence. The picture, taken thus, is not seen as occurring in the same system as when one sees it as the representation of a particular position of the circle between the lines. Or rather, taken thus, it is a part of a different calculus. The rules that govern variables are not the same as those that govern their particular values. "How do you know he is in the room?" "Because I put him in and there is no way he can get out." Then your knowledge of the general fact that he is somewhere in the room has the same multiplicity as that reason. Let us take the particular case of the general state of affairs of the cross being between the end-lines.

~)~(--111

)( I I

xl 1--1---;x~1

Each of these cases, for instance, has its own individuality. Is there any way in which this individuality enters into the sense of the general sentence? Obviously not. 'Being between the lines, or the walls' seems something simple and the particular positions (both the visual appearances and the

1.57

positions established by measurement) seem quite independent of it. That is, when we talk about the individual (seen) positions we appear to be talking about something quite different from the topic of the general proposition. There is one calculus containing our general characterization and another containing the disjunction. If we say that the cross is between the lines we don't have any disjunction ready to take the place of the general proposition. If we consider a general proposition like "the circle is in the square" it appears time and again that the assignment of a position in the square is not (at least so far as visual space is concerned) a more precise specification of the statement that the circle is in the square any more than a statement of the colour of a material is a more precise specification of a statement of its hardness. - Rather, "in the square" appears a complete specification which in itself does not admit of any more precise description. Now of course the statements about the circle are not related to each other like the statements about colour and hardness, and yet that feeling is not baseless. The grammatical rules for the terms of the general proposition must contain the multiplicity of possible particular cases provided for by the proposition. What isn't contained in the rules isn't provided for.

All these patterns might be the same state of affairs distorted. (Imagine the two white strips and the middle black strip as elastic.) Does fa's following from (x). fx mean that a is mentioned in

(x).Ex? Yes, if the general proposition is meant in such a way that its verification consists in an enumeration.

If I say "there is a black circle in the square", it always seems to me that here again I have something simple in mind, and don't have to think of different possible positions or sizes of the circle. And yet one may say: if there is a circle in the square, it must be somewhere and have some size. But in any case there cannot be any question of my thinking in advance of all the possible positions and sizes. - It is rather that in the first proposition I seem to put them through a kind of sieve so that "circle in a square" corresponds to a single impression, which doesn't take any account of the where etc., as if it were (against all appearance) something only physically, and not logically, connected with the first state of affairs. The point of the expression "sieve" is this. If I look at a landscape or something similar through a glass which transmits only the distinction between brightness and darkness and not the distinctions between colours, such a glass can be called a sieve; and if one thinks of the square as being looked at through a glass which transmits only the distinction "circle in the square or not in the square" and no distinction between positions or sizes of the circle, here too we might speak of a sieve. I would like to say that in the proposition "there is a circle in the square" the particular positions are not mentioned at all. In the picture I don't see the position, I disregard it, as if the distances from the sides of the square were elastic and their lengths of no account. Indeed, can't the patch actually be moving in the square? Isn't that just a special case of being in the square? So in that case it wouldn't be true that the patch has to be in a particular position in the square if it is there at all. I want to say that the patch seems to have a relation to the edge that is independent of its distance. - Almost as if I were using a

r

geometry in which there is no such thing as distance, but only inside and outside. Looked at in this way, there is no doubt that the two pictures

G D and

6 The proposition "The circle is in the square" is not a di.[}unction of cases.

are the same. If I say the patch is in the square, I know - and must knowthat it may have various possible positions. I know too that I uldn't give a definite number of all such positions. I do not co . h" . know in advance how many positions "I could d"lstmg':ls And trying it out won't tell me what I want to know here elther. The darkness veiling the possible positions etc. is the current logical situation, just as dim lighting is a particular sort of lighting.

By itself the proposition "The patch is in the square" does no more than hold the patch in the square, as it were; it is only in this way that it limits the patch's freedom; within the square it allows it complete freedom. The proposition constructs a frame that limits the freedom of the patch but within the frame it leaves it free, that is, it has nothing to do with its position. For that to be so the proposition must have the logical nature of the frame (like a box enclosing the patch). And so it has, because I could explain the proposition to someone and set out the possibilities, quite independently of whether such a proposition is true or not, independently of a fact.

Here it always seems as if we can't quite get an overall view of a logical form because we don't know how many or what possible positions there are for the patch in the square. But on the other hand we do know, because we aren't surprised by any of them when they turn up.

"Wherever the patch is in the square ... " means "as long as it is in the square ... " and here all that is meant is the freedom (lack of restraint) in the square, not a set of positions. j

!

Of course "position of the circle in this square" isn't a concept which particular positions fall under as objects. You couldn't discover objects and ascertain that they were positions of the circle in the square which you didn't know about beforehand.

Of course between this freedom and the totality of possibilities, there is a logical similarity (formal analogy), and that is why the same words are often used in the two cases ("all", "every", etc.).

Incidentally, the centre and other special positions in the circle are quite analogous to the primary colours on the colour scale. (This comparison might be pursued with profit.)

"No degrees of brightness below this one hurt my eyes." Test the type of generality.

Space is as it were a single possibility; it doesn't consist of several possibilities.

"All points on this surface are white." How do you verify that? - then I will know what it means.

So if I hear that the book is somewhere on the table, and then find it in a particular position, it isn't possible for me to be surprised and say "oh, I didn't know that there was this position"; and yet I hadn't foreseen this particular position i.e. envisaged it in

.z60

.z6I

L

advance as a particular possibility. It is physical, not logical possibilities that take me by surprise!

that case isn't the disjunction of all those N positions the very

roposition that A is somewhere between Band C? . P But what are these N pictures really like? It is clear that a picture roust not be visually discernible from its immediate successor, or the transition will be discontinuous. The positions whose succession I see as a continuous transition are positions which are not in visual space.

But what is the difference between "the book is somewhere on the table" and "the event will occur sometime in the future?" Obviously the difference is that in the one case we have a sure method of verifying whether the book is on the table, while in the other case there is no such method. If a particular event were supposed to occur at one of the infinitely many bisections of a line, or better, if it were supposed to occur when we cut the line at a single point, not further specified, and then waited a minute at that point, that statement would be as senseless as the one about the infinite future.

How is the extension of the concept "lying between" determined? Because it has to be laid down in advance what possibilities belong to this concept. As I say, it cannot be a surprise that I call that too "lying between". Or: how can the rules for the expression "lie between" be given when I can't enumerate the cases of lying between? Of course that itself must be a characteristic of the meaning of the expression.

Suppose I stated a disjunction of so many positions that it was impossible for me to see a single position as distinct from all those given; would that disjunction be the general proposition (3x). fx ? ~ouldn't it be a kind of pedantry to continue to refuse to recognIze the disjunction as the general proposition? Or is there an essential distinction" and is the disjunction totally unlike the general proposition?

Indeed if we wanted to explain the word to someone we wouldn't try to do so by indicating all particular instances, but by showing hi~ one or two such instances and intimating in some way that it wasn't a question of the particular case. It is not only that the enumeration of positions is unnecessary: in the nature of things there can be no question of such an enumeration here.

What so strikes us is that the one proposition is so complicated and the other so simple. Or is the simple one only an abbreviation for the more complicated one? What then is the criterion for the general proposition, for the circle's being in the square? Either, nothing that has anything to do with a set of positions (or sizes) or something that deals with a finite number of such positions.

Saying "The circle is either between the two lines or here" (where "here" is a place between the lines) obviously means no more than "The circle is between the two lines", and the rider "or here" is superfluous. You will say: the "here" is already included in the "somewhere". But that is strange, since it isn't mentioned in it.

If one says that the patch A is somewhere between the limits B and C, isn't it obviously possible to describe or portray a number of positions of A between Band C in such a way that I see the succession of all the positions as a continuous transition? And in

There is a particular difficulty when the signs don't appear to say what the thought grasps, or the words don't say what the thought appears to grasp.

26;

262

L

1I

As when we say "this theorem holds of all numbers" and think ~hat in our thought we have comprehended all numbers like apples In a box.

7 The inadequacy of the Frege-Russell notation for generality

But now it might be asked: how can I know in advance which propos~t~ons entail this general proposition, if I can't specify the proposltlons ?

The real difficulty lies in the concept of "(3n)" and in general of "(3x)". The original source of this notation is the expression of our word-language :"There is a ... with such and such properties". And here what replaces the dots is something like "book from my library" or "thing (body) in this room", "word in this letter", etc. We think of objects that we can go through one after the other. As so often happens a process of sublimation turned this form into "there is an object such that ... " and here too people imagined originally the objects of the world as like 'objects' in the room (the tables, chairs, books, etc.), although it is clear that in many cases the grammar of this "(3x), etc." is not at all the same as the grammar of the primitive case which serves as a paradigm. The discrepancy between the original picture and the one to which the notation is now applied becomes particularly palpable when a proposition like "there are two circles in this square" is rendered as "there is no object that has the property of being a circle in this square without being the circle a or the circle b" or "there are not three objects that have the property of being a circle in this square". The proposition "there are only two things that are circles in this square" (construed on the model of the proposition "there are only two men who have climbed this mountain") sounds crazy, with good reason. That is to say, nothing is gained by forcing the proposition "there are two circles in this square" into that form; it only helps to conceal that we haven't cleared up the grammar of the proposition. But at the same time the Russellian notation here gives an appearance of exactitude which makes people believe the problems are solved by putting the proposition into the Russellian form. (This is no less dangerous than using the word "probably" without further investigation into the use of the word in this particular case. For understandable reasons the word

But can one say "We can't say which propositions entail this proposition"? That sounds like: we don't know. But of course that isn't how it is. I can indeed say, and say in advance, propositions that entail it. "Only not all of them." But that just has no meaning. There is just the general proposition and particular propositions (not the particular propositions). But the general proposition does not enumerate particular propositions. In that case what characte~izes it as general, and what shows that it doesn't simply comprIse the particular propositions. we are speaking of in this particular case? It cannot be characterized by its instantiations, because however many.we enumerate, it could still be mistaken for the product of the cIted. cases. Its generality, therefore, lies in a property (a grammatical property) of the variables.

L

J

"probably", too, is connected with an idea of exactitude.) "One of the four legs of this table doesn't hold","There are Englishmen with black hair", "There is a speck on this wall" "The two pots have the same weight", "There are the same number of words on each of the two pages". In all these cases in the Russellian notation the "(3 ... ) ... " is used, and each time with a different grammar. The point I want to make is that nothing much is gained by translating such a sentence from word-language into Russellian notation. It makes sense to say "write down any cardinal number" but not "write down all cardinal numbers". "There is a circle in the square" [(3x).fx)] makes sense, but not -3x. - fx: "all circles are in the square.;' "There is a red circle on a background of a different colour" makes sense, but not "there isn't a backgroundcolour other than red that doesn't have a red circle on it." "In this square there is a black circle". If this proposition has the form "(3x).x is a black circle in a square" what sort of thing is it that has the property of being a black circle (and so can also have the property of not being a black circle)? Is it a place in the square? But then there is no proposition "(x). x is a black ... " On the other hand the proposition could mean "There is a speck in the square that is a black circle". How is that proposition verified? Well, we take the different specks in the square in turn and investigate whether they are quite black and circular. But what kind of proposition is "There isn't a speck in the square"? For if in the former case the 'x' in '(3x)' meant 'speck in the square', then though "(3x). fx" is a possible proposition both "(3x)" and "-(3x)" are not. Or again, I might ask: what sort of thing is it that has (or does not have) the property of being a speck in the square? And if we can say "There is a speck in the square" does it then also make sense to say "All specks are in the square"? All which?

266

Ordinary language says "In this square there is a red circle"; the Russellian notation says "There is an object which is a red circle in this square". That form of expression is obviously modelled on "There is a substance which shines in the dark" "There is a circle in this square which is red". - Perhaps even the expression "there is" is misleading. "There is" really means the same as "Among these circles there is one ... " or " ... there exists " one .... So if we go as far as we can in the direction of the Russellian mode of expression and say "In this square there is a place where there is a red circle", that really means, among these places there is one where ... etc. (In logic the most difficult standpoint is that of sound common sense. For in order to justify its view it demands the whole truth; it will not help by the slightest concession or construction.) The correct expression of this sort of generality is therefore the expression of ordinary language "There is a circle in the square", which simply leaves the position of the circle open (leaves it undecided). ("Undecided" is a correct expression, since there just has not been any decision.)

8

Incidentally, we really could introduce a notation for (3x). cpx in which it was replaced by a sign "cpr v cps v cpt ... " which could then be used in calculation like a logical sum; but we would have to provide rules for reconverting t~is. not~ti~n at any ~ime,~nto the "(3x) . cpx" notation and thus dl.stmgulshmg the. slgn ~a v cpb v cpc ... " from the sign for a loglcal sum. The pomt of thlS notation could simply be to enable us to calculate more easily with (3x) . cpx in certain cases.

Criticism of my former view of generaliry

My view about general propositions was that (3x) . cpx is a logical sum and that though its terms aren't enumerated here, they are capable of being enumerated (from the dictionary and the grammar of language). For if they can't be enumerated we don't have a logical sum. (A rule, perhaps, for the constructio'1 of logical sums).

If I am right, there is no concept "pure colour"; the proposition "A's colour is a pure colour" simply means "A is red, or yellow, or blue, or green". "This hat belongs either to A or B or C" is not the same proposition as "This hat belongs to a person in this room" even when in fact only A, Band C are in the room, for that itself is something that has to be added. - "On this surface there are two pure colours" means: on this surface there is red and yellow, or red and green, or ... etc. If this means I can't say "there are 4 pure colours", still the pure colours and the number 4 are somehow connected with each other and that must express itself in some way. - For instance, I may say "on this surface I see 4 colours: yellow, blue, red, green".

Of course, the explanation of (3x). cpx as a logical sum and of (x). cpx as a logical product is indefensible. It went with an incorrect notion of logical analysis in that I thought that some day the logical product for a particular (x). cpx would be found. - Of course it is correct that (3x). cpx behaves in some ways like a logical sum and (x). cpx like a product; indeed for one use of words "all" and "some" myoid explanation is correct, - for instance for "aU the primary colours occur in this picture" or "all the notes of the C major scale occur in this theme". But for cases like "all men die before they are 200 years old" my explanation is not correct. The way in which (3x). cpx behaves like a logical sum is expressed by its following from cpa and from cpavlcpb, i.e. in the rules

The generality notation of our ordinary language grasps the logical form even more superficially than I earlier believed. In this respect it is comparable with the subject-predicate form.

(3x). cpx: cpa. = . cpa and (3x) . cpx: cpa v cpb. = . cpa v cpb

Generality is as ambiguous as the subject-predicate form.

From these rules Russell's fundamental laws follow as tautologies: cpx. ::l . (3z) . cpz cpx v cpy. ::l .(3z). cpz

There are as many different "aIls" as there are different "ones". So it is no use using the word "all" for clarification unless we know its grammar in this particular case.

For (3x). cpx we need also the rules: (3x). cpx v ~x. = .(3x). cpx. v .(3x). ~x (3x,y) cpx.~y. v .(3x). cpx. ~x. = .(3x). cpx :(3x).~x. Every such rule is an expression of the analogy between (3x). cpx and a logical sum.

268

L

9 The explanation oj generality by examples

sign for "some", since he fixes his attention on something different; but what matters is the system of rules governing the signs, and it isn't seeing the signs in a particular manner that is the essence of understanding.

Let us think how we explain the concept plant. We show someone several objects and say they are plants; then he points to another object and asks "is that a plant too?" and we reply "yes, that too" etc. I would once have said that he has now seen in what he has been shown the concept 'plant' - the common element - and that he does not see the examples used in the explanation in the same way when he sees the concept in them as when he views them just as representatives of a particular shape and colour or the like. (Just as I also used to say that when he understands variables as variables he sees something in them which he doesn't see in the sign for the particular case). But the notion of "seeing in" is taken from the case in which I see a figure like 1111 differently "phrased". In that case, I really do see different figures, but in a different sense; and what these have in common, apart from their similarity, is their being caused by the same physical pattern. But this explanation cannot be applied without further ado to the case of the understanding of a variable or of the examples illustrating the concept "plant". For suppose we really had seen something in them that we don't see in plants that are shown only for their own sake, the question remains whether this, or any other, picture can entitle us to apply them as variables. I might have shown someone the plants by way of explanation and given him in addition a drug causing him to see the examples in the special way. (Just as it would be possible that a drunken man might always see a group like 1111 as III I). And this would give the explanation of the concept in an unambiguous manner, and the specimens exhibited and the accompanying gestures would communicate to anyone who understood just this picture. But that is not the way it is. - It may well be true that someone who sees a sign like 111111 as a numeral for 6 sees it differently (sees something different in it) from someone who views it only as a

It would be possible to say "now I don't see it as a rose, but as a plant". Or "now I see it only as a rose, and no longer as this rose". "I see the patch merely in the square and no longer in a specific position. "

The mental process of understanding is of no interest to us (any more than the mental process of an intuition). "Still, there's no doubt that someone who understands the examples as arbitrary cases chosen to illustrate the concept doesn't understand the same as a man who regards them as a definitely bounded enumeration." Quite right, but what does the first man understand that the second doesn't? Well, in the things he is shown he sees only examples to illustrate certain features; he doesn't think that I am showing him the things for their own sake as well. I would like to call the one class "logically bounded" and the other "logically unbounded". Yes, but is it really true that he sees only these features in the things? In a leaf, say, does he see only what is common to all leaves? That would be as if he saw everything else blank like an uncompleted form with the essential features ready printed. (But the function "f( ... )" is just such a form.) But what sort of a process is it when someone shows me several different things as examples of a concept to get me to see what is common to them, and when I look for it and then actually see it? He may draw my attention to what is common. - But by doing this

2.7 0

27 1

L

does he make me see the object differently? Perhaps so; for surely I may take a special look at one of the parts, when otherwise I would have seen the whole with equal clarity. But this seeing is not the understanding of the concept. For what we see isn't something with an empty argument place. One might also ask: Does a man who regards the sign" III ... " as a sign for the concept of number (in contrast with" III" to denote 3) see the first group of lines differently from the second? Even if he does see it differently (perhaps, as it were, more blurred) does he see there anything like the essence of the concept of number? Wouldn't that mean that he would actually have to be unable to distinguish "III ... " and "1111 ... " from each other ? (As indeed he would, if I had given him some drug that made him see the concept.) For if I say: by givingus a few examples he makes us see the common element in them and disregard the rest, that really means that the rest falls into the background, as it were becomes paler (or altogether disappears - why not ?) and "the common element", say the oval shape, remains alone in the foreground. But that isn't the way it is. Apart from anything else, the multiplicity of examples would be no more than a mechanical device, and once I had seen what I was supposed to, I could see it in a single example too. (As indeed '(3x). fx' itself contains only one example.) So it is the rules governing the example that make it an example. But by now at any rate, if someone says to me something like "make an egg shape" the bare concept word without any illustration suffices to make itself understood (and the past history of this understanding is of no interest to us): and I do not want to say that when I understand the command (and the word "egg") I see

the concept of an egg before my mind's eye. When I make an application of the concept "egg" or "plant" there certainly isn't some general picture in front of my mind before I do so, and when I hear the word "plant" it isn't that there comes before my mind a picture of a certain object which I then describe as a plant. No, I make the application as it were spontaneously. Still, in the case of certain applications I might say "No, I didn't mean that by 'plant' ", or, "Yes, I meant that too". But does that mean that these pictures came before my mind and that mentally I expressly rejected and admitted them? And yet that is what it looks like, when I say: "Yes, I meant all those things, but not that." But one might then ask: "But did you foresee all those cases?" and then the answer might be "yes" or "no, but I imagined there must be something between this form and that one" or the like. But commonly at that moment I did not draw any bounds, and they can only be produced in a roundabout way after reflection. For instance, I say "Bring me a flower about so big"; he brings one and I say : Yes, that is the size I meant. Perhaps I do remember a picture which came before my mind, but it isn't that that makes the flower that has been brought acceptable. What I am doing is making an application of the picture, and the application was not anticipated. The only thing of interest to us is the exact relationship between the example and the behaviour that accords with it. The example is the point of departure for further calculation. Examples are decent signs, not rubbish or hocus-pocus. The only thing that interests us is the geometry of the mechanism. (That means, the grammar of its description.) But how does it come out in our rules, that the instances of fx

we are dealing with are not essentially closed classes? - Only indeed in the generality of the general rule. - How does it come out that they don't have the same significance for the calculus as a closed group of primitive signs (like the names of the 6 basic colours)? How else could it come out except in the rules given for them? - Suppose that in some game I am allowed to help myself to as many pieces as I like of a certain kind, while only a limited number of another kind is available; or suppose a game is unbounded in time but spatially bounded, or something similar. The case is exactly the same. The distinction between the two different types of piece in the game must be laid down in the rules; they will say about the one type that you can take as many pieces as you want of that kind. And I mustn't look for another more restrictive expression of that rule. That means that the expression for the unboundedness of the particular instances in question will be a general expression; there cannot be some other expression in which the other unconsidered instances appear in some shadowy way. It is clear that I do not recognize any logical sum as a definition of the proposition "the cross is between the lines". And that says everything that is to be said.

There is one thing I always want to say to clarify the distinction between instances that are offered as examples for a concept and instances that make up a definite closed group in the grammar. Suppose, after explaining "a, b, c, d are books", someone says "Now bring me a book". If the person brings a book which isn't one of the ones shown him he can still be said to have acted correctly in accordance with the rule given. But if what had been said was "a, b, c, d, are my books. - Bring me one of my books", it would have been incorrect to bring a different one and he would have been told "I told you that a, b, c, d are my books". In the first case it isn't against the rule to bring an object other than those named,

274

in the second case it is. But if in the order you named only a, b, c, and d, and yet you regarded the behaviour fee) as obeying the order, doesn't that mean that by F(a, b, c, d, ... ) you meant F(a, b, c, d, e) after all ? Again, how are these orders distinct from each other if the same thing obeys both of them? - But f(g) too would have been in accordance with the order and not only fee). Right, then your first order must have meant F(a, b, c, d, e, g) etc. Whatever you bring me is something I could have included in a disjunction. So if we construct the disjunction of all the cases we actually use, how would it differ syntactically from the general proposition? For we can't say: by the fact that the general proposition is also made true by r (which doesn't occur in the disjunction), because that doesn't distinguish the general proposition from a disjunction which contains r. (And every other similar answer too is impossible.) But it will make sense to say: FCa, b, c, d, e) is the disjunction of all the cases we have actually used, but there are also other cases (we won't of course, mention any) that make true the general proposition "F(a, b, c, d, ... )". And here of course we can't put the general proposition in place of F(a, b, c, d, e,). It is, by the way, a very important fact that the parenthesis in the previous paragraph "and every other similar answer too is impossible" is senseless, because though you can give as instances of a generalization different particular cases, you can't give different variables because the variables r, s, t don't differ in their meaning.

Of course one couldn't say that when we do f( d) we don't obey f(3) in the same way as we obey a disjunction containing f( d), because f(3) = f(3) v f( d). If you give someone the order "bring me some plant or other, or this one" (giving him a picture of it), he will simply discard the picture and say to himself "since anyone will do, the picture doesn't matter". By contrast, we won't simply

discard the picture if we are given it plus five others and the order to bring one of these six plants. (So what matters is which disjunction contains the particular command.) And you wouldn't be guided in the same way by the order "f(a) v feb) v f(c)" as by the order "f(3)" (= f(3) v f(c», even if in each case you do f(c). - The picture f( c) sinks into f(3). (It is no good sitting in a boat, if you and it are under water and sinking). Someone may be inclined to say: "Suppose you do f( c) on the command f(3); in that case f( c) might h?ve been expressly permitted and then how would the general command have differed from a disjunction?" - But if the permission had occurred in a disjunction with the general sentence, you couldn't have appealed to it. So is this how it is: "bring me a flower" can never be replaced by an order of the form "bring me a or b or c", but must always be "bring me a or b or c or some other flower" ? But why does the general sentence behave so indeterminately when every case which actually occurs is something I could have described in advance? But even that seems to me not to get to the heart of the matter; because W'hat matters, I believe, isn't really the infinity of the possibilities, but a kind of indeterminacy. Indeed, if I were asked how many possibilities a circle in the visual field has of being within a particular square, I could neither name a finite number, nor say that there were infinitely many (as in a Euclidean plane). Here, although we don't ever come to an end, the series isn't endless in the way in which I I, ~, ~ + I I is. Rather, no end to which we come is really the end; that is, I could always say: I don't understand why these should be all the possibilities. - And doesn't that just mean that it is senseless to speak of "all the possibilities"? So enumeration doesn't touch the concepts "plant" and "egg" at all.

And although we say that we could always have forseen f(a) as a possible particular execution of the order, still we didn't in fact ever do so. - But even if I do foresee the possibility f(a) and expressly include it in my order, it gets lost beside the general proposition, because I can see from the general proposition itself that this particular case is permitted; it isn't just from its being expressly permitted in the order that I see this. If the general proposition is there, the addition of the particular case isn't any extra use to me (that is, it doesn't make the command more explicit). Indeed it was only the general proposition that gave me the justification for placing this particular case beside it. What my whole argument is aiming at, is that someone might believe that the addition of the particular case supersedes the - as it were blurred - generality of the proposition, that you could say "we don't need it any more, now we have the particular case." Yes, but say I admit that the reason I put in the particular case is that it agrees with the general proposition! Or suppose I admit that I recognize that f(a) is a particular case of f(3)! For I can't say: that just means that f(3) is a disjunction with f(a) as one of its terms; for if that is so, the disjunction must be capable of being stated, and f(3) must be defined as a disjunction. There would be no difficulty in giving such a definition, but it wouldn't correspond to the use of f(3) that we have in mind. It isn't that the disjunction always leaves something over; it is that it just doesn't touch the essential thing in generality, and even if it is added to it it depends on the general proposition for its justification. First I command f(3); he obeys the order and does f(a). Then I think that I could just as well have given him the command "f(3) v f(a)". (For I knew in advance that f(a) obeyed the order f(3) and to command him f(3) v f(a) would come to the same.) In that case when he obeyed the order he would have been acting on the disjunction "do something or f( a)". Andifhe obeys the order

for that would mean that a complete definition was possible. That would be the disjunction which would make the addition "v f(3)" as it were ridiculous, since it would only be the enumerated instances which concerned us. But according to our idea of f(3), the stipulation that f(a) is a case of f(3) is not an incomplete definition of f(3); it is not a definition of £(3) at all. That means that I don't approximate to the sense of f(3) by multiplying the number of cases in the disjunction; though the disjunction of the cases v f(3) is equivalent to f(3), it is never equivalent to the disjunction of the cases alone; it is a totally different proposition.

by doing f(a) isn't it immaterial what else is disjoined with f(a)? If he does f( a) in any case, the order is obeyed whatever the alternative is. I w~uld ~lso like to say: in grammar nothing is supplementary, no stlpUlatlOns come after others, everything is there simultaneously. Thus I can't even say that I first gave the command f(3) and only later realised that f( a) was a case of f(3); at all events my order was and remained f(3) and I added f(a) to it in the knowledge that f(a) was in accordance with f(3). And the stipulation that f(a) is in accordance with f(3) presupposes the sense that belongs to the proposition f(3) if,it is taken as an independent unit and not defined as replaceable by a disjunction. And my proposition "at all events my order was and remained f(3) etc." only means that I didn't replace the general order by a disjunction. Suppose I give the order p v f(a), and the addressee doesn't clearly understand the first part of the order but does understand that the order goes" ... v f(a)". He might then do f(a) and say "I know for certain that I've obeyed the command, even though I didn't understand the first part". And that too is how I imagine it when I say that the other alternative doesn't matter. But in that case he didn't obey the order that was given, but simply treated it as "f(a)!" One might ask: if someone does f(a) at the command "f(3) v f(a)" is he obeying the order because (i.e. in so far as) the order is of the form ~ v f(a), or because f(3) v f(a) = f(3)? If you understand f(3) and therefore know that f(3) v f(a) = f(3), then by doing f(a) you are obeyingf(3)evenifI write it "f(3) v f(a)"because you can see none the less that f(a) is a case of f(3). And now someone might object: if you see that Fa is a case of F(3) that just means that f(a) is contained disjunctively in f(3), and therefore that f(3) is defined fry means of f(a). The remaining parts of the disjunction he will have to say - don't concern me because the terms I see are the only ones I now need. - By explaining 'that f(a) is an instance of f(3)' "you have said no more than that f(a) occurs in f(3) alongside certain other terms." - But that is precisely what we don't mean. It isn't as if our stipulation was an incomplete definition of f(3);

What is said about an enumeration of individual cases cannot ever be a roundabout explanation of generality. But can I give the rules of entailment that hold in this case? How do I know that (3x) . fx does follow from fa? After all I can't give all the propositions from which it follows. - But that isn't necessary; if (3x) . fx follows from fa, that at any rate was something that could be known in advance of any particular experience, and stated in the grammar. I said "in advance of any experience it was possible to know and to state in the grammar that (3x).fx follows from fa". But it should have been: '(3x).fx follows from fa' is not a proposition (empirical proposition) of the language to which '(3x). fx' and 'fa' belong; it is a rule laid down in their grammar.

279

L

I

1 10

way without really enumerating or enumerating to a determined limit.

The law of a series "And so on"

The dots in " I + I + I + I . . . " are just the four dots: a sign, for which it must be possible to give certain rules. (The sam::: rules, in fact, as for the sign "and so on ad inf.".) This sign does in a manner ape enumeration, but it isn't an enumeration. And that means that the rules governing it don't totally agree with those which govern an enumeration; they agree only up to a point.

We can of course set up a rule for the use of the variables, and the fact that in order to do so we need the same kind of variable does not make it pleonastic. For if we didn't use it, then the variable would be defined by the rules, and we don't assume that it can be defined, or that it must be defined (for sooner or later definitions come to an end).

There is no third thing between the particular enumeration and the general sign.

This means only that - e.g. - the variable "x2 " is not an abbreviation (say for a logical sum), and that in our thought too there is only a sign for this multiplicity.

Of course the natural numbers have only been written down up to a certain highest point, let's say 10 10 • Now what constitutes the possibility of writing down numbers that have not yet been written down? How odd is this feeling that they are all somewhere already in existence! (Frege said that before it was drawn a construction line was in a certain sense already there.)

For suppose I had enumerated 7 particular instances and said "but their logical sum isn't the general proposition" that still wouldn't be enough; and I want to say further that no other number of instances yields the general proposition either. But in this rider once again I seem to go through an enumeration, in a kind of shadowy manner if not in actuality. But that is not the way it is, because the words that occur in the rider are quite different from the numerals.

The difficulty here is to fight off the thought that possibility is a kind of shadowy reality. In the rules for the variable a a variable b may occur and so may particular numerals; but not any totality of numbers.

"But how can I forbid a particular numeral to be inserted in such and such a place? I surely can't foresee what number someone will want to insert, so that I can forbid it". You can forbid it when it comes. - But here we are already speaking of the general concept of number!

But now it seems as if this involved denying the existence of something in logic: perhaps generality itself, or what the dots indicate; whatever is incomplete (loose, capable of further extension) in the number series. And of course we may not and cannot deny the existence of anything. So how does this indeterminacy find expression? Roughly thus: if we introduce numbers substitutible for the variable a, we don't say of any of them that it is the last, or the highest.

But what makes a sign an expression of infinity? What gives the peculiar character that belongs to what we call infinite? I believe that it is like the case of a sign for an enormous number. For the characteristic of the infinite, conceived in this way, is its enormous size. But there isn't anything that is an enumeration and yet not an enumeration; a generality that enumerates in a cloudy kind of 280

i

,~'

281

J

But suppose someone asked us after the explanation of a form of calculation "and is 103 the last sign I can use?" What are we to answer? "No, it isn't the last" or "there isn't a last?" Mustn't I ask him in turn "If it isn't the last, what would come next?" And if he then says "104" I should say "Quite right, you can continue the series yourself". Of an end to the possibility, I cannot speak at all. (In philosophy the one thing we must guard against is waffle. A rule that can be applied in practice is always in order.) It is clear that we can follow a rule like Ia, ~, ~ + 1 I. I mean by really following the rule for constructing it without previously being able to write down the series. In that case it's the same as if I were to begin a series with a number like 1 and then say "now add 7, multiply by 5, take the square root of the result, and always apply this complex operation once again to the result". (That

would be the rule

I1, ~, v(~ + 7)·51·)

The expression "and so on" is nothing but the expression "and so on" (nothing, that is, but a sign in a calculus which can't do more than have meaning via the rules that hold of it; which can't say more than it shows). That is, the expression "and so on" does not harbour a secret power by which the series is continued without being continued. Of course it doesn't contain that, you'll say, but still it contains the meaning of infinite continuation. But we might ask: how does it happen that someone who now applies the general rule to a further number is still following this rule? How does it happen that no further rule was necessary

282

to allow him to apply the general rule to this case in spite of the fact that this case was not mentioned in the general rule? And so we are puzzled that we can't bridge over this abyss between the individual numbers and the general proposition. "Can one imagine an empty space?" (Surprisingly, this is where this question belongs.) It is one of the most deep rooted mistakes of philosophy to see possibility as a shadow of reality. But on th~ other hand it can't be an error; not even if one calls the proposition such a shadow. Here again, of course, there is a danger of falling into a positivism, of a kind which deserves a special name, and hence of course must be an error. For we must avoid accepting party lines or particular views of things; we must not disown anything that anyone has ever said on the topic, except where he himself had a particular view or theory. For the sign "and so on", or some sign corresponding to it, is essential if we are to indicate endlessness - through the rules, of course, that govern such a sign. That is to say, we can distinguish the limited series "1, 1 + 1, 1 + 1 + 1" from the series "1 1 + 1 I + I + 1 and so on". And this last sign and its use is no les~ essen~ tial for the calculus than any other. What troubles me is that the "and so on" apparently has to occur also in the rules for the sign "and so on". For instance, 1, I + I and so on . = . 1, 1 + 1, I + 1 + 1 and so on, and so on. But then isn't this simply the old point that we can describe language only from the outside? So that we can't expect by describing language to penetrate to depths deeper than language

I

, itself reveals: for it is by means of language that we describe language. We might say: there's no occasion to be afraid of our using the expression "and so on" in a way that transcends the finite. Moreover, the distinctive part of the grammar of "and so on" can't consist in rules connecting "and so on" with particular numerals (not "the particular numerals") - for these rules in turn mention some bit of a series - but in rules connecting "and so on" with "and so on". The possibility of introducing further numbers. The difficulty seems to be that the numbers I've in fact introduced aren't a group that is essential and yet there is nothing to indicate that they are an arbitrary collection: Out of all numbers just those numbers that happen to have been written down. (As ifI had all the pieces of a game in a box and a chance selection from the box on the table beside it. Or, as if one lot of numerals was traced in ink, while all of them are as it were drawn faintly in advance.) But apart from the ones we happen to have used we have only the general form. Isn't it here, by the way, - odd as it may sound - that the distinction between numerals and numbers comes? Suppose, for example, I say "By 'cardinal number' I mean whatever results from I by continued addition of I". The word "continued" doesn't represent a nebulous continuation of I, I + I, I + I + I ; on the contrary the sign" I, I + I, I + I + I . . . " is to be taken as perfectly exact; governed by definite rules which are different from those for" I, I + I, I + I + I " , and not a substitute for a series "which cannot be written down".

In other words: we calculate with the sign" I, I + I, I + I + I ... " just as with the numerals, but in accordance with different rules. But what is it then that we imagine? What is the mistake we make? What kind of thing do we take the sign" I, I + I . • • " to be ? That is : where does what we think we see in this sign realfy occur? Something like when I say "he counted I, 2, 3, 4 and so on up to 1000", where it would also be possible really to write down all the numbers. What do we see" I, I + I, I + I + I . . . . " as ? As an inexact form of expression. The dots are like extra numerals indistinctly visible. It is as if we stopped writing numerals, because after all we can't write them all down, but as if they are there all right in a kind of box. Again, it is something like when I sing only the first notes of a melody distinctly, and then merely hint at the rest and let it taper off into nothing. (Or when in writing one writes only a few letters of a word distinctly and ends with an unarticulated line.) In all such cases the 'indistinctfy' has a 'distinctfy' corresponding to it. I once said that there couldn't be both numbers and and the concept of number. And that is quite correct, if it means that a variable doesn't have the same relation to a number as the concept apple has to an apple (or the concept sword to Nothung). On the other hand, a number-variable is not a numeral. But I also wanted to say that the concept of number couldn't be given independently of the numbers, and that isn't true. A number-variable is independent of particular numbers in the sense that there does exist a calculus with a class of our numerals and without the general number-variable. In that calculus, of course, not all the rules which hold of our numerals will be valid, but those numerals will correspond to ours in the way that the draughtsmen in draughts correspond to those in losing draughts. What I am opposing is the view that the infinite number series is

j

something given concerning which there are both particular number theorems and also general theorems about all numbers of the series; so that the arithmetical calculus wouldn't be complete if it didn't contain the general theorems about cardinal numbers, i.e. general equations of the form a + (b + c) = (a + b) + c. Whereas even 1/3 = o· 3 belongs to a different calculus from 1/3 = 0·3. And similarly a general sign-rule (e.g. a recursive definition) that holds for I, (I) + 1,((1) + I) + 1,(((1) + I) + I) + I, and so on is something different from a particular definition. The general rule adds to the number calculus something extra, without which it would have been no less complete than the arithmetic of the number series I, 2, 3, 4, 5. The question also arises: where is the concept of number (or of cardinal number) indispensable? Number, in contrast to what? II I, ~, ~ + 1 I, perhaps, in contrast to I 5, ~ y'~ I etc. - For if! really do introduce such a sign (like I I, ~, ~ + 1 \) and don't just take it along as a luxury, then I must do something with it, i.e. use it in a calculus, and then it loses its solitary splendour and occurs in a system of signs coordinated with it. You will perhaps say: but surely "cardinal number" is contrasted with "rational number", "real number", etc. But this distinction is a distinction between the rules (the rules of the appropriate game) - not a distinction between positions on the chessboardnot a distinction demanding different coordinated words in the same calculus. We say "this theorem is proved for all cardinal numbers". But let us just see how the concept of cardinal numbers enters into the proof. Only because 1 and the operation ~ + 1 are spoken of in the proof - not in contrast to anything the rational numbers have. So if we use the concept-word "cardinal number" to describe the proof in prose, we see - don't we? - that no concept corresponds to that word.

286

The expressions "the cardinal numbers", "the real numbers", are extraordinarily misleading except where they are used to help specify particular numbers, as in "the cardinal numbers from 1 to 100", etc. There is no such thing as "the cardinal numbers", but only "cardinal numbers" and the concept, the form "cardinal number". Now we say "the number of the cardinal numbers is smaller than the number of the real numbers" and we imagine that we could perhaps write the two series side by side (if only we weren't weak humans) and then the one series would end in endlessness, whereas the other would go on beyond it into the actual infinite. But this is all nonsense. If we can talk of a relationship which can be called by analogy "greater" and "smaller", it can only be a relationship between the forms "cardinal number" and "real number". I learn what a series is by having it explained to me and only to the extent that it is explained to me. A finite series is explained to me by examples of the type I, 2, 3, 4, and infinite one by signs of the type" I, 2, 3, 4, and so on" or "I, 2, 3, 4 ... " It is important that I can understand (see) the rule of projection without having it in front of me in a general notation. I can detect a general rule in the series t, i, ~, 146 - of course I can detect any number of others too, but still I can detect a particular one, and that means that this series was somehow for me the expression of that one rule. If you have "intuitively" understood the law of a series, e.g. the series m, so that you are able to construct an arbitrary term men), then you've completely understood the law, just as well as anything like an algebraic formulation could convey it. That is, no such formulation can now make you understand it better, and therefore to that extent no such formulation is any more rigorous, although it may of course be easier to take in. We are inclined to believe that the notation that gives a series by writing down a few terms plus the sign "and so on" is essentially inexact, by contrast with the specification of the general term.

'or

Here we forget that the general term is specified by reference to a basic series which cannot in turn be described by a general term. Thus 2n + I is the general term of the odd numbers, if n ranges over the cardinal numbers, but it would be nonsense to say that n was the general term of the series of cardinal numbers. If you want to define that series, you can't do it by specifying "the general term n", but of course only by a definition like" I, I + I, I + I + I and so on". And of course there is no essential difference between that series and " I , I + I + I, I + I + I + I + I and so on", which I could just as well have taken as the basic series (so that then the general term of the cardinal number series would have been

ten -

I).) (3x). ([)x: -- (3x, y). ([)x. ([)y (3x,y).([)x.([)y: --(3x,y,z).([)x.([)y.([)Z (3x, y, Z). ([)x. ([)y.([)Z: --(3x, y, z, u).([)x.([)y.([)z.([)U

"How would we now go about writing the general form of such propositions? The question manifestly has a good sense. For if I write down only a few such propositions as examples, you understand what the essential element in these propositions is meant to be." Well, in that case the row of examples is already a notation: for understanding the series consists in our applying the symbol, and distinguishing it from others in the same system, e.g. from (3x). ([)X (3x, y,z).([)x.([)y.([)z. (3x,y,z,u,v).([)x.([)y.([)z.([)u.([)V But why shouldn't we write the general term of the first series thus: (3Xl' ... Xn).n~; ([)X :(3x\ ... xn + 1). n::+1 ([)X?l Is this notation inexact? It isn't supposed by itself to make anything graphic; all that matters are the rules for its use, the system in which it is used. The scruples attaching to it date from a train of thought which was concerned with the number of primitive signs in the calculus of Principia Mathematica. I. Perhaps Wittgenstein inadvertently omitted a negation sign before the second quantifier. (Trs.)

288

III FOUNDATIONS OF MATHEMATICS II

The comparison between mathematics and a game What are we taking away from mathematics when we say it is only a game (or: it is a game)? A game, in contrast to what? - What are we awarding to mathematics if we say it isn't a game, its propositions have a sense? The sense outside the proposition. What concern is it of ours? Where does it manifest itself and what can we do with it? (To the question "what is the sense of this proposition ?" the answer is a proposition.) ("But a mathematical proposition does express a thought." What thought? -.) Can it be expressed by another proposition? Or only by this proposition? - Or not at all? In that case it is no concern of ours. Do you simply want to distinguish mathematical propositions from other constructions, such as hypotheses? You are right to do so: there is no doubt that there is a distinction. If you want to say that mathematics is played like chess or patience, and the point of it is like winning or coming out, that is manifestly incorrect. If you say that the mental processes accompanying the use of mathematical symbols are different from those accompanying chess, I wouldn't know what to say about that. In chess there are some positions that are impossible although each individual piece is in a permissible position. (E.g. if all the pawns are still in their initial position, but a bishop is already in

play.) But one could imagine a game in which a record was kept of the number of moves from the beginning of the game and then there would be certain positions which could not occur after n moves and yet one could not read off from a position by itself whether or not it was a possible nth position.

different from a person who wanted to find out how many

What we do in games must correspond to what we do in calculating. (I mean: it's there that the correspondence must be, or again, that's the way that the two must be correlated with each other.)

Well, we might want to compare this with a game like tennis. In tennis the player makes a particular movement which causes the ball to travel in a particular way, and we can view his hitting the ball either as an experiment, leading to the discovery of a particular truth, or else as a stroke with the sole purpose of winning the game. But this comparison wouldn't fit, because we don't regard a move in chess as an experiment (though that too we might do); we regard it as a step in a calculation.

2I X

8

is? It will be said: the one wanted to find out a truth, but the other

did not want to do anything of the sort.

Is mathematics about signs on paper? No more than chess is about wooden pieces. When we talk about the sense of mathematical propositions, or what they are about, we are using a false picture. Here too, I mean, it looks as if there are inessential, arbitrary signs which have an essential element in common, namely ¢e sense.

Someone might perhaps say: In the arithmetical game we do indeed do the multiplication 2 I X 8,

Since mathematics is a calculus and hence isn't really about anything, there isn't any metamathematics.

168

but the equation 21 x 8 = 168 doesn't occur in the game. But isn't that a superficial distinction? And why shouldn't we multiply (and of course divide) in such a way that the equations were written down as equations?

What is the relation between a chess problem and a game of chess? - It is clear that chess problems correspond to arithmetical problems, indeed that they are arithmetical problems. The following would be an example of an arithmetical game: We write down a four-figure number at random, e.g. 7368; we are to get as near to this number as possible by multiplying the numbers 7, 3, 6, 8 with each other in any order. The players calculate with pencil and paper, and the person who comes nearest to the number 73 68 in the smallest number of steps wins. (Many mathematical puzzles, incidentally, can be turned into games of this kind.)

So one can only object that in the game the equation is not a proposition. But what does that mean? How does it become a proposition? What must be added to it to make it a proposition? - Isn't it a matter of the use of the equation (or of the multiplication)? - And it is certainly a piece of mathematics when it is used in the transition from one proposition to another. And thus the specific difference between mathematics and a game gets linked up with the concept of proposition (not 'mathematical proposition') and thereby loses its actuality for us.

Suppose a human being had been taught arithmetic only for use in an arithmetical game: would he have learnt something different from a person who learns arithmetic for its ordinary use? If he multiplies 21 by 8 in the game and gets 168, does he do something

But one could say that the real distinction lay in the fact that in the game there is no room for affirmation and negation. For

I

L

instance, there is multiplication and ZI x 8 = 148 would be a false move, hut "( ZIX 8 = 148)", which is a correct arithmetical proposition, would have no business in our game.

problem and therefore a "problem" in the sense of physics; a chess problem is a mathematical problem and so a "problem" in a different sense, a mathematical sense.

(Here we may remind ourselves that in elementary schools they never work with inequations. The children are only asked to carry out multiplications correctly and never - or hardly ever - asked to prove an inequation.)

In the debate between "formalism" and "contentful mathematics" what does each side assert? This dispute is so like the one between realism and idealism in that it will soon have become obsolete, for example, and in that both parties make unjust assertions at variance with their day-to-day practice.

When I work out Z 1 X 8 in our game the steps in the calculation, at least, are the same as when I do it in order to solve a practical problem (and we could make room in a game for inequations also). But my attitude to the sum in other respects differs in the two cases. Now the question is: can we say of someone playing the game who reaches the position "ZI x 8 = 168" that he has found out that ZI x 8 is 168? What does he lack? I think the only thing missing is an application for the sum. Calling arithmetic a game is no more and no less wrong than calling moving chessmen according to chess-rules a game; for that might be a calculation too. So we should say: No, the word "arithmetic" is not the name of a game. (That too of course is trivial) - But the meaning of the word "arithmetic" can be clarified by bringing out the relationship between arithmetic and an arithmetical game, or between a chess problem and the game of chess. But in doing so it is essential to recognize that the relationship is not the same as that between a tennis problem and the game of tennis. By "tennis problem" I mean something like the problem of returning a ball in a particular direction in given circumstances. (A billiard problem would perhaps be a clearer case.) A billiard problem isn't a mathematical problem (although its solution may be an application of mathematics). A billiard problem is a physical

Arithmetic isn't a game, it wouldn't occur to anyone to include arithmetic in a list of games played by human beings. What constitutes winning and losing in a game (or success in patience)? It isn't of course, just the winning position. A special rule is needed to lay down who is the winner. ("Draughts" and "losing draughts" differ only in this rule.) Now is the rule which says "The one who first has his pieces in the other one's half is the winner" a statement? How would it be verified? How do I know if someone has won? Because he is pleased, or something of the kind? Really what the rule says is: you must try to get your pieces as soon as possible, etc. In this form the rule connects the game with life. And we could imagine that in an elementary school in which one of the subjects taught was chess the teacher would react to a pupil's bad moves in exactly the same way as to a sum worked out wrongly. I would almost like to say: It is true that in the game there isn't any "true" and "false" but then in arithmetic there isn't any "winning" and "losing".

L II 11

;1

1

!\

'I

I

I

I

_J

dispute between formalism and intuitionism, etc. People cannot separate the importance, the consequences, the application of a fact from the fact itself; they can't separate the description of a thing from the description of its importance.

I once said that is was imaginable that wars might be fought on a kind of huge chessboard according to the rules of chess. But if everything really went simply according to the rules of chess, then you wouldn't need a battlefield for the war, it could be played on an ordinary board; and then it wouldn't be a war in the ordinary sense. But you really could imagine a battle conducted in accordance with the rules of chess - if, say, the "bishop" could fight with the "queen" only when his position in relation to her was such that he would be allowed to "take" her in chess.

We are always being told that a mathematician works by instinct (or that he doesn't proceed mechanically like a chessplayer or the like), but we aren't told what that's supposed to have to do with the nature of mathematics. If such a psychological phenomenon does playa part in mathematics we need to know how far we can speak about mathematics with complete exactitude, and how far we can only speak with the indeterminacy we must use in speaking of instincts etc.

Could we imagine a game of chess being played (i.e. a complete set of chess moves being carried out) in such different surroundings that what happened wasn't something we could call the playing of a game? Certainly, it might be a case of the two participants collaborating to solve a problem. (And we could easily construct a case on these lines in which such a task would have a utility).

Time and again I would like to say: What I check is the account books of mathematicians; their mental processes, joys, depressions and instincts as they go about their business may be important in other connections, but they are no concern of mine.

The rule about winning and losing really just makes a distinction between two poles. It is not concerned with what later happens to the winner (or loser) - whether, for instance, the loser has to pay anything. (And similarly, the thought occurs, with "right" and "wrong" in sums.) In logic the same thing keeps happening as happened in the dispute about the nature of definition. If someone says that a definition is concerned only with signs and does no more than substitute one sign for another, people resist and say that that isn't all a definition does, or that there are different kinds of definition and the interesting and important ones aren't the mere "verbal definitions" . They think, that is, that if you make definition out to be a mere substitution rule for signs you take away its significance and importance. But the significance of a definition lies in its application, in its importance for life. The same thing is happening today in the

1'\

294

1

12

There is no metamathematics.

No calculus can decide a philosophical problem. A calculus cannot give us information about the foundations of mathematics. So there can't be any "leading problems" of mathematical logic, if those are supposed to be problems whose solution would at long last give us the right to do arithmetic as we do. We can't wait for the lucky chance of the solution of a mathematical problem. I said earlier "calculus is not a mathematical concept"; in other words, the word "calculus" is not a chesspiece that belongs to mathematics. There is no need for it to occur in mathematics. - If it is used in a calculus nonetheless, that doesn't make the calculus into a metacalculus; in such a case the word is just a chessman like all the others. Logic isn't metamathematics either; that is, work within the logical calculus can't bring to light essential truths about mathematics. Cf. here the "decision problem" and similar topics in modern mathematical logic. (Through Russell and Whitehead, especially Whitehead, there entered philosophy a false exactitude that is the worst enemy of real exactitude. At the bottom of this there lies the erroneous opinion that a calculus could be the mathematical foundation of mathematics.) Number is not at all a "fundamental mathematical concept"l. 1. According to Dr. C. Lewy Wittgenstein wrote in the margin of F. P. Ramsey's copy of the Tractatus at 6.02: "Number is the fundamental idea of calculus and must be introduced as such." This was, Lewy thinks, in the year 1923. See Mind, July 1967, p. 422.

There are so many calculations in which numbers aren't mentioned. So far as concerns arithmetic, what we are willing to call numbers is more or less arbitrary. For the rest, what we have to do is to describe the calculus - say of cardinal numbers - that is, we must give its rules and by doing so we lay the foundations of arithmetic. Teach it to us, and then you have laid its foundations. (Hilbert sets up rules of a particular calculus as rules of metamathematics. ) A system's being based on first principles is not the same as its being developed from them. It makes a difference whether it is like a house resting on its lowest walls or like a celestial body floating free in space which we have begun to build beneath although we might have built anywhere else. Logic and mathematics are not based on axioms, any more than a group is based on the elements and operations that define it. The idea that they are involves the error of treating the intuitiveness, the self-evidence, of the fundamental propositions as a criterion for correctness in logic. A foundation that stands on nothing is a bad foundation. (p . q) v (p. - q) v (- p . q) v (- p. - q . ): That is my tautology, and then I go on to say that every "proposition of logic" can be brought into this form in accordance with specified rules. But that means the same as: can be derived from it. This would take us as far as the Russellian method of demonstration and all we add to it

is that this initial form is not itself an independent proposition, and that like all other "laws of logic" it has the property that p . Log = p, P V Log = Log.

13 Proofs of Relet!ance

It is indeed the essence of a "logical law" that when it is conjoined with any proposition it yields that proposition. We might even begin Russell's calculus with definitions like

If we prove that a problem can be solved, the concept "solution" must occur somewhere in the proof. (There must be something in the mechanism corresponding to the concept.) But the concept cannot have an external description as its proxy; it must be genuinely spelt out.

p~p:q.=.q

p : p v q. = . p, etc. The only proof of the provability of a proposition is a proof of the proposition itself. But there is something we might call a proof of relevance: an example would be a proof convincing me that I can verify the equation 17 x 38 = 456 before I have actually done so. Well, how is it that I know that I can check 17 x 38 = 456, whereas I perhaps wouldn't know, merely by looking, whether I could check an expression in the integral calculus? Obviously, it is because I know that the equation is constructed in accordance with a definite rule and because I know the kind of connection between the rule for the solution of the sum and the way in which the proposition is put together. In that case a proof of relevance would be something like a formulation of the general method of doing things like multiplication sums, enabling us to recognize the general form of the propositions it makes it possible to check. In that case I can say I recognise that this method will verify the equation without having actually carried out the verification. When we speak of proofs of relevance (and other similar mathematical entities) it always looks as if in addition to the particular series of operations called proofs of relevance, we had a quite definite inclusive concept of such proofs or of mathematical proof in general; but in fact the word is applied with many different, more or less related, meanings. (Like words such as "people", "king", "religion", etc.; cf Spengler.) Just think of the role of examples in the explanation of such words. If I want to explain what I mean by "proof", I will have to point to examples of proofs,

just as when explaining the word "apple" I point to apples. The definition of the word "proof" is in the same case as the definition of the word "number". I can define the expression "cardinal number" by pointing to examples of cardinal numbers; indeed instead of the expression I can actually use the sign" I, 2,3,4, and so on ad inf". I can define the word "number" too by pointing to various kinds of number; but when I do so I am not circumscribing the concept "number" as definitely as I previously circumscribed the concept cardinal number, unless I want to say that it is only the things at present called numbers that constitute the concept "number", in which case we can't say of any new construction that it constructs a kind of number. But the way we wantto use the word "proof" in is one in which it isn't simply defined by a disjunction of proofs currently in use; we want to use it in cases of which at present we "can't have any idea". To the extent that the concept of proof is sharplY circumscribed, it is only through particular proofs, or through series of proofs (like the number series), and we must keep that in mind if we want to speak absolutely precisely about proofs of relevance, of consistency etc. We can say: A proof of relevance alters the calculus containing the proposition to which it refers. It cannot justify a calculus containing the proposition, in the sense in which carrying out the multiplication 17 x 23 justifies the writing down of the equation 17 x 23 = 391. Not, that is, unless we expressly give the word "justify" that meaning. But in that case we mustn't believe that if mathematics lacks this justification, it is in some more general and widely established sense illegitimate or suspicious. (That would be like someone wanting to say: "the use of the expression 'pile of stones' is fundamentally illegitimate, until we have laid down officially how many stones make a pile." Such a stipulation would modify the use of the word "pile" but it wouldn't "justify" it in any generally recognized sense; and if such an official definition were given, it wouldn't mean that the use earlier made of the word would be stigmatized as incorrect.)

A B

The proof of the verifiability of 17 x 23 = 391 is not a "proof" in the same sense of the word as the proof of the equation itself. (A cobbler heels, a doctor heals: both ... ) We grasp the verifiability of the equation from its proof somewhat as we grasp the verifiability of the proposition "the points A and B are not separated by a turn of the spiral" from the figure. And we see that the proposition stating verifiability isn't a "proposition" in the same sense as the one whose verifiability is asserted. Here again, one can only say: look at the proof, and you will see what is proved here, what gets called "the proposition proved". Can one say that at each step of a proof we need a new insight? (The individuality of numbers.) Something of the following sort: ifI am given a general (variable) rule, I must recognize each time afresh that this rule may be applied here too (that it holds for this case too). No act of foresight can absolve me from this act of insight. Since the form in which the rule is applied is in fact a new one at every step. But it is not a matter of an act of insight, but of an act of decision. What I called a proof of relevance does not climb the ladder to its proposition - since that requires that you pass every rungbut only shows that the ladder leads in the direction of that propo-

sition. (There are no surrogates in logic). Neither is an arrow that points the direction a surrogate for going through all the stages towards a particular goal.

14

Consistenry proofs

Something tells me that a contradiction in the axioms of a system can't really do any harm until it is revealed. We think of a hidden contradiction as like a hidden illness which does harm even though (and perhaps precisely because) it doesn't show itself in an obvious way. But two rules in a game which in a particular instance contradict each other are perfectly in order until the case turns up, and it's only then that it becomes necessary to make a decision between them by a further rule. Mathematicians nowadays make so much fuss about proofs of the consistency of axioms. I have the feeling that if there were a contradiction in the axioms of a system it wouldn't be such a great misfortune. Nothing easier than to remove it. "We may not use a system of axioms before its consistency has been proved." "In the rules of the game no contradictions may occur." Why not? "Because then one wouldn't know how to play." But how does it happen that our reaction to a contradiction is a doubt? We don't h:Jve any reaction to a contradiction. We can only say: if it's really meant like that (if the contradiction is supposed to be there) I don't understand it. Or: it isn't something I've learnt. I don't understand the sign. J haven't learnt what I am to do with it, whether it is a command, etc. Suppose someone wanted to add to the usual axioms of arithmetic the equation 2 x 2 = 5. Of course that would mean that the sign of equality had changed its meaning, i.e. that there would now be different rules for the equals-sign.

30 3

302 I

\

1

I

~

supposition, then the supposition must make sense. - But what does it mean to make that supposition? It isn't making a supposition that goes against natural history, like the supposition that a lion has two tails. - It isn't making a supposition that goes against an ascertained fact. What it means is supposing a rule; and there's nothing against that except that it contradicts another rule, and for that reason I drop it. Suppose that in the proof there occurs the following drawing to represent a straight line bifurcating. There is nothing absurd (contradictory) in that unless we have made some stipulation that it contradicts.

If I inferred "I cannot use it as a substitution sign" that would mean that its grammar no longer fitted the grammar of the word "substitute" ("substitution sign", etc.). For the word "can" in that proposition doesn't indicate a physical (physiological, psychological) possibility. "The rules many not contradict each other" is like "negation, when doubled, may not yield a negation". That is, it is part of the grammar of the word "rule" that if "p" is a rule, "p. ~ p" is not a rule.

_____======---

That means we could also say: the rules may contradict each other, if the rules for the use of the word "rule" are different - if the word "rule" has a different meaning.

If a contradiction is found later on, that means that hitherto the rules have not been clear and unambiguous. So the contradiction doesn't matter, because we can now get rid of it by enunciating a rule.

Here too we cannot give any foundation (except a biological or historical one or something of the kind); all we can do is to establish the agreement, or disagreement between the rules for certain words, and say that these words are used with these rules.

In a system with a clearly set out grammar there are no hidden contradictions, because such a system must include the rule which makes the contradiction is discernible. A contradiction can only be hidden in the sense that it is in the higgledy-piggledy zone of the rules, in the unorganized part of the grammar; and there it doesn't matter since it can be removed by organizing the grammar.

It cannot be shown, proved, that these rules can be used as the rules of this activity. Except by showing that the grammar of the description of the activity fits the rules.

Why may not the rules contradict one another? Because other· wise they wouldn't be rules.

"In the rules there mustn't be a contradiction" looks like an instruction: "In a clock the hand mustn't be loose on the shaft." We expect a reason: because otherwise ... But in the first case the reason would have to be: because otherwise it wouldn't be a set of rules. Once again we have a grammatical structure that cannot be given a logical foundation.

I

In the indirect proof that a straight line can have only one continuation through a certain point we make the supposition that a straight line could have two continuations. - If we make that

I

I,

11

~

1

15 Justifying arithmetic and preparing it for its applications (Russell, Ramsry)

One always has an aversion to giving arithmetic a foundation by saying something about its application. It appears firmly enough grounded in itself. And that of course derives from the fact that arithmetic is its own application. You could say: why bother to limit the application of arithmetic, that takes care of itself. (I can make a knife without bothering about what kinds of materials I will have cut with it; that will show soon enough.) What speaks against our demarcating a region of application is the feeling that we can understand arithmetic without having any such region in mind. Or put it like this: our instinct rebels against anything that isn't restricted to an analysis of the thoughts already before us. You could say arithmetic is a kind of geometry; i.e. what in geometry are constructions on paper in arithmetic are calculations (on paper). You could say, it is a more general kind of geometry. It is always a question of whether and how far it's possible to represent the most general form of the application of arithmetic. And here the strange thing is that in a certain sense it doesn't seem to be needed. And if in fact it isn't needed, then it's also impossible. The general form of its application seems to be represented by the fact that nothing is said about it. (And if that's a possible representation, then it is also the right one.) The point of the remark that arithmetic is a kind of geometry is simply that arithmetical constructions are autonomous like

geometrical ones and hence so to speak themselves guarantee their applicability. For it must be possible to say of geometry too that it is its own application. (In the sense in which we can speak of lines which are possible and lines which are actually drawn we can also speak of possible and actually represented numbers.)

11111111111 ------------I

I

1

That is an arithmetical construction, and in a somewhat extended sense also a geometrical one. Suppose I wish to use this calculation to solve the following problem: if I have I I apples and want to share them among some people in such a way that each is given 3 apples how many people can there be? The calculation supplies me with the answer 3. Now suppose I were to go through the whole process of sharing and at the end 4 people each had 3 apples ~n their hands. Would I then say that the computation gave a wrong result? Of course not. And that of course means only that the computation was not an experiment. It might look as though the mathematical computation entitled us to make a prediction, say, that I could give three people their share and there will be two apples left over. But that isn't so. What justifies us in making this prediction is an hypothesis of physics, which lies outside the calculation. The calculation is only a study of logical forms, of structures, and of itself can't yield anything new. If 3 strokes on the paper are the sign for the number 3, then you can say the number 3 is to be applied in our language in the way in which the 3 strokes can be applied. I said "One difficulty in the Fregean theory is the generality of the words 'Concept' and 'Object'. For, even if you can count

tables, tones, vibrations and thoughts, it is difficult to bracket them all together." I But what does "you can count them" mean? What it means is that it makes sense to apply the cardinal numbers to them. But if we know that, if we know these grammatical rules, why do we need to rack our brains about the other grammatical rules when we are only concerned to justify the application of cardinal arithmetic? It isn't difficult "to bracket them all together"; so far as is necessary for the present purpose they are already bracketed together. But (as we all know well) arithmetic isn't at all concerned about this application. Its applicability takes care of itself. Hence so far as the foundations of arithmetic are concerned all the anxious searching for distinctions between subject-predicate forms, and constructing functions 'in extension' (Ramsey) is a waste of time. The equation 4 apples + 4 apples = 8 apples is a substitution rule which I use if instead of substituting the sign "8" for the sign "4 + 4", I substitute the sign "8 apples" for the sign "4 + 4 apples." But we must beware of thinking that "4 apples + 4 apples = 8 apples" is the concrete equation and 4 + 4 = 8 the abstract proposition of which the former is only a special case, so that the arithmetic of apples, though much less general than the truly general arithmetic, is valid in its own restricted domain (for apples). There isn't any "arithmetic of apples", because the equation 4 apples + 4 apples = 8 apples is not a proposition about apples. We may say that in this equation the word "apples" has no reference. (And we can always say this about a sign in a rule which helps to determine its meaning.)

How can we make preparations for the reception of something that may happen to exist - in the sense in which Russell and Ramsey always wanted to do this? We get logic ready for the existence of many-placed relations, or for the existence of an infinite number of objects, or the like. Well, we can make preparations for the existence of a thing: e.g. I may make a casket for jewellery which may be made some time or other - But in this case I can say what the situation must be - what the situation is - for which I am preparing. It is no more difficult to describe the situation now than after it has already occurred; even, if it never occurs at all. (Solution of mathematical problems). But what Russell and Ramsey are making preparations for is a possible grammar. On the one hand we think that the nature of the functions and of the arguments that are counted in mathematics is part of its business. But we don't want to let ourselves be tied down to the functions now known to us, and we don't know whether people will ever discover a function with 100 argument places; and so we have to make preparations and construct a function to get everything ready for a Ioo-place relation in case one turns up. - But what does "a Ioo-place relation turns up (or exists)" mean at all? What concept do we have of one? Or of a z-place relation for that matter? - As an example of a z-place relation we give something like the relation between father and son. But what is the significance of this example for the further logical treatment of z-place relations? Instead of every "aRb" are we now to imagine "a is the father of b"? - If not, is this example or any example essential? Doesn't this example have the same role as an example in arithmetic, when I use 3 rows of 6 apples to explain 3 x 6 = 18 to somebody? Here it is a matter of our concept of application. - We have an image of an engine which first runs idle, and then works a machine. But what does the application add to the calculation? Does it

1.

Philosophical Remarks, p.

119.

introduce a new calculus? In that case it isn't any longer the same calculation. Or does it give it substance in some sense which is essential to mathematics (logic)? If so, how can we abstract from the application at all, even only temporarily? No, calculation with apples is essentially the same as calculation with lines or numbers. A machine is an extension of an engine, an application is not in the same sense an extension of a calculation. Suppose that, in order to give an example, I say "love is a z-place relation" - am I saying anything about love? Of course not. I am giving a rule for the use of the word "love" and I mean perhaps that we use this word in such and such a way. Yet we do have the feeling that when we allude to the z-place relation 'love' we put meaning into the husk of the calculus of relations. - Imagine a geometrical demonstration carried out using the cylinder of a lamp instead of a drawing or analytical symbols. How far is this an application of geometry? Does the use of the glass cylinder in the lamp enter into the geometrical thought? And does the use of word "love" in a declaration of love enter into my discussions of z-place relations? We are concerned with different uses or meanings of the word "application". "Division is an application of multiplication"; "the lamp is an application of the glass cylinder"; "the calculation is applied to these apples". At this point we can say: arithmetic is its own application. The calculus is its own application. In arithmetic we cannot make preparations for a grammatical application. For if arithmetic is only a game, its application too is only a game, and either the same game (in which case it takes us no further) or a different game - and in that case we could play it in pure arithmetic also. I

;1 r.j

II :1 jl I' I I

310

So if the logician says that he has made preparations in arithmetic for the possible existence of 6-place relations, we may ask him: when what you have prepared finds its application, what will be added to it? A new calculus? - but that's something you haven't provided. Or something which doesn't affect the calculus? - then it doesn't interest us, and the calculus you have shown us is application enough. What is incorrect is the idea that the application of a calculus in the grammar of real language correlates it to a reality or gives it a reality that it did not have before. Here as so often in this area the mistake lies not in believing something false, but in looking in the direction of a misleading analogy. So what happens when the 6-place relation is found? Is it like the discovery of a metal that has the desired (and previously described) properties (the right specific weight, strength, etc.)? No; what is discovered is a word that we in fact use in our language as we used, say, the letter R. "Yes, but this word has meaning, and 'R' has none. So now we see that something can correspond to 'R'." But the meaning of the word does not consist in something's corresponding to it, except in a case like that of a name and what it names; but in our case the bearer of the name is merely an extension of the calculus, of the language. And it is not like saying "this story really happened, it was not pure fiction". This is all connected with the false concept of logical analysis that Russell, Ramsey and I used to have, according to which we are writing for an ultimate logical analysis of facts, like a chemical analysis of compounds - an analysis which will enable us really

to discover a 7-place relation, like an element that really has the specific weight 7. Grammar is for us a pure calculus (not the application of a calculus to reality). "How can we make preparations for something which mayor may not exist?" means: how can we hope to make an a priori construction to cope with all possible results while basing arithmetic upon a logic in which we are still waiting for the results of an analysis of our propositions in particular cases? - One wants to say: "we don't know whether it may not turn out that there are no functions with 4 argument places, or that there are only 100 arguments that can significantly be inserted into functions of one variable. Suppose, for example (the supposition does appear possible) that there is only one four-place function F and 4 arguments a, b, c, d; does it make sense in that case to say '2 + 2 = 4' since there aren't any functions to accomplish the division into 2 and 2?" So now, one says to oneself, we will make provision for all possible cases. But of course that has no meaning. On the one hand the calculus doesn't make provision for possible existence; it constructs for itself all the existence that it needs. On the other hand what look like hypothetical assumptions about the logical elements (the logical structure) of the world are merely specifications of elements in a calculus; and of course you can make these in such a way that the calculus does not contain any 2 + 2. Suppose we make preparations for the existence of 100 objects by introducing 100 names and a calculus to go with them. Then let us suppose 100 objects are really discovered. What happens now that the names have objects correlated with them which weren't correlated with them before? Does the calculus change? - What has the correlation got to do with it at all ? Does it make it acquire more reality? Or did the calculus previously belong only to mathematics, and now to logic as well? - What sort of questions are "are there 3-place relations", "are there 1000 objects"? How is

it to be decided? - But surely it is a fact that we can specify a 2-place relation, say love, and a 3-place one, say jealousy, but perhaps not a z7-place one! - But what does "to specify a z-place relation" mean? It sounds as if we could point to a thing and say "you see, that is the kind of thing" (the kind of thing we described earlier). But nothing of that kind takes place (the comparison with pointing is altogether wrong). "The relation of jealousy cannot be reduced to z-place relationships" sounds like "alcohol cannot be decomposed into water plus a solid substance". Is that something that is part of the nature of jealousy? (Let's not forget: the proposition "A is jealous ofB because of C" is no more and no less reducible than the proposition "A is not jealous ofB because of C".) What is pointed to is, say, the group of people A, Band C. - "But suppose that living beings at first knew only plane surfaces, but none the less developed a 3-dimensional geometry, and that they suddenly became acquainted with 3-dimensional space!" Would this alter their geometry, would it become richer in content? - "Isn't this the way it is? Suppose at some time I had made arbitrary rules for myself prohibiting me from moving in my room in certain directions where there were no physical hindrances to get in my way; and then suppose the physical conditions changed, say furniture was put in the room, in such a way as to force me to move in accordance with the rules which I had originally imposed on myself arbitrarily. Thus, while the 3-dimensional calculus was only a game, there weren't yet three dimensions in reality because the x, y, z belonged to the rules only because I had so decided; but now that we have linked them up to the real 3 dimensions, no other movements are possible for them." But that is pure fiction. There isn't any question here of a connection with reality which keeps grammar on the rails. The "connection of language with reality", by means of ostensive definitions and the like, doesn't make the grammar inevitable or provide a justification for the grammar. The grammar remains a free-floating calculus which can only be extended and never supported. The "connection with

reality" merely extends language, it doesn't force anything on it. We speak of discovering a 27-place relation but on the one hand no discovery can force me to use the sign or the calculus for a 27-place relation, and on the other hand I can describe the operation of the calculus itself simply by using this notation. When it looks in logic as if we are discussing several different universes (as with Ramsey), in reality we are considering different games. The definition of a "universe" in a case like Ramsey's would simply be a definition like

16 RamsI!)" s theory of identity

Ramsey's theory of identity makes the mistake that would be made by someone who said that you could use a painting as a mirror as well, even if only for a single posture. If we say this we overlook that what is essential to a mirror is precisely that you can infer from it the posture of a body in front of it, whereas in the-case of the painting you have to know that the postures tally before you can construe the picture as a mirror image.

Def

(3x) . CflX = Cfla v Cflb v CflC v Cfld. If Dirichlet's conception of function has a strict sense, it must be expressed in a definition that uses the table to define the functionsigns as equivalent. Ramsey defines 1 x = y as ( Cfle) . Cfle X == CfleY But according to the explanations he gives of his function-sign " Cfle" (Cfle)· Cfle X == Cflex is the statement: "every sentence is equivalent to itself." (Cfle) . CfleX == CfleY is the statement: "every sentence is equivalent to every sentence." So all he has achieved by his definition is what is laid down by the two definitions X=X. X=y.

Def Def

Tautology Contradiction

(Here the word "tautology" can be replaced by any arbitrary tautology, and similarly with "contradiction"). So far all that has happened is that definitions have been given of the two distinct signs x = x and x = y. These definitions could of course be replaced by two sets of definitions, e.g. a=a } b= b = Taut. c=c I.

314

a= b } b= c = Contr. c=a

F. P. Ramsey, The Foundations of Mathematics, London 1931, p. 53.

315

For the signs "fa" "fb" "fc" are no more function and argument than the words "Co(rn)", "Co(al)" and "Co(It)" are. (Here it makes no difference whether or not the "arguments" "rn", "al", "It" are used elsewhere as words). (So it is hard to see what purpose the definitions can have except to mislead us.) To begin with, the sign "(3x) .fx" has no meaning; because here the rules for functions in the old sense of the word don't hold at all. According to them a definition like fa = ... would be nonsense. If no explicit definition is given for it, the sign "(3x). fx" can only be understood as a rebus in which the signs have some kind of spurious meaning. Each of the signs "a = a", "a = c", etc. in the definitions (a = a). Def • Taut. etc. is a word. Moreover, the purpose of the introduction of functions in extension was to analyse propositions about infinite extensions, and it fails of this purpose when a function in extension is introduced by a list of definitions.

But then Ramsey writes: "(3x, y). x oF y",

i.e.

"(3x, y). - (x = y)" -

but he has no right to: for what does the "x = y" mean in this expression? It is neither the sign "x = y" used in the definition above, nor of course the "x = x" in the preceding definition. So it is a sign that is still unexplained. Moreover to see the futility of these definitions, you should read them (as an unbiased person would) as follows: I permit the sign "Taut", whose use we know, to be replaced by the sign "a = a" or "b = b", etc.; and the sign "Contr." (" - Taut.") to be replaced by the sign "a = b" or "a = c", etc. From which, incidentally, it follows that (a = b) = (c = d) = (a oF a) = etc.! It goes without saying that an identity sign defined like that has no resemblance to the one we use to express a substitution rule. Of course I can go on to define "(3x, y).x oF y", say as a oF a. v. a oF b. V • b oF c. V • a oF c; but this definition is pure humbug and I should have written straightaway (3x, y). x oF y. r;:f. Taut. (That is, I would be given the sign on the left side as a new unnecessary - sign for "Taut. ") For we mustn't forget that according to the definitions "a = a", "a = b", etc. are independent signs, no more connected with each other than the signs "Taut." and "Contr." themselves. What is in question here is whether functions in extension are any use; because Ramsey's explanation of the identity sign is just such a specification by extension. Now what exactly is the specification of a function by its extension? Obviously, it is a group of definitions, e.g.

i

fa= p Def. fb= q Def. fc = r Def. I

These definitions permit us to substitute for the known propositions "p", "q", "r" the signs "fa" "fb" "fc". To say that these three definitions determine the function f(~) is either to say nothing, or to say the same as the three definitions say.

31 6

I

There is a temptation to regard the form of an equation as the form of tautologies and contradictions, because it looks as if one can say that x = x is self-evidently true and x = y self-evidently false. The comparison between x = x and a tautology is of course better than that between x = y and a contradiction, because all correct (and "significant") equations of mathematics are actually of the form x = y. We might call x = x a degenerate equation (Ramsey quite correctly called tautologies and contradictions degenerate propositions) and indeed a correct degenerate equation (the limiting case of an equation). For we use expressions of the form x = x like correct equations, and when we do so we are fully conscious that we are dealing with degenerate equations. In geometrical proofs there are propositions in the same case, such as "the angle IX is equal to the angle ~ , the angle 'Y is equal to itself ... " At this point the objection might be made that correct equations of the form x = y must be tautologies, and incorrect ones contradictions, because it must be possible to prove a correct equation

317

by transforming each side of it until an identity of the form x = x is reached. But although the original equation is shown to be correct by this process, and to that extent the identity x = x is the goal of the transformation, it is not its goal in the sense that the purpose of the transformation is to give the equation its correct formlike bending a crooked object straight; it is not that the equation at long last achieves its perfect form in the identity. So we can't say: a correct equation is reallY an identity. It just isn't an identity.

17

The concept of the application of arithmetic 1 (mathematics)

If we say "it must be essential to mathematics that it can be applied" we mean that its applicabiliry isn't the kind of thing I mean of a piece of wood when I say "I will be able to find many applications for it". Geometry isn't the science (natural science) of geometric planes, lines and points, as opposed to some other science of gross physical lines, stripes and surfaces and their properties. The relation between geometry and propositions of practical life, about stripes, colour boundaries, edges and corners, etc. isn't that the things geometry speaks of, though ideal edges and corners, resemble those spoken of in practical propositions; it is the relation between those propositions and their grammar. Applied geometry is the grammar of statements about spatial objects. The relation between what is called a geometrical line and a boundary between two colours isn't like the relation between something fine and something coarse, but like the relation between possibility and actuality. (Think of the notion of possibility as a shadow of actuality.) You can describe a circular surface divided diametrically into 8 congruent parts, but it is senseless to give such a description of an elliptical surface. And that contains all that geometry says in this connexion about circular and elliptical surfaces. (A proposition based on a wrong calculation (such as "he cut a 3-metre board into 4 one metre parts") is nonsensical, and that throws light on what is meant by "making sense" and "meaning something by a proposition".)

I. The section does not mention arithmetic. It may be conjectured that it was never completed. (Ed.)

What about the proposition "the sum of the angles of a triangle is 180 degrees"? At all events you can't tell by looking at it that it is a proposition of syntax. The proposition "corresponding angles are equal" means that if they don't appear equal when they are measured I will treat the measurement as incorrect; and "the sum of the angles of a triangle is 180 degrees" means that if it doesn't appear to be 180 degrees when they are measured I will assume there has been a mistake in the measurement. So the proposition is a postulate about the method of describing facts, and therefore a proposition of syntax.

IV ON CARDINAL NUMBERS 18

Kinds of cardinal number What are numbers? - What numerals signify; an investigation of what they signify is an investigation of the grammar of numerals. What we are looking for is not a definition of the concept of number, but an exposition of the grammar of the word "number" and of the numerals. The reason why there are infinitely many cardinal numbers, is that we construct this infinite system and call it the system of cardinal numbers. There is also a number system "I, 2, 3, 4, 5, many" and even a system "1,2,3,4,5". Why shouldn't I call that too a system of cardinal numbers (a finite one)? It is clear that the axiom of infinity is not what Russell took it for; it is neither a proposition of logic, nor - as it stands - a proposition of physics. Perhaps the calculus to which it belongs, transplanted into quite different surroundings (with a quite different "interpretation"), might somewhere find a practical application; I do not know. One might say of logical concepts (e.g. of the, or a, concept of infinity) that their essence proves their existence.

(Frege would still have said: "perhaps there are people who have not got beyond the first five in their acquaintance with the series of cardinal numbers (and see the rest of the series only in an indeterminate form or something of the kind), but this series exists independently of us". Does chess exist independently of us, or not? -) Here is a very interesting question about the position of the concept of number in logic: what happens to the concept of

number if a society has no numerals, but for counting, calculating, etc. uses exclusively an abacus like a Russian abacus? (Nothing would be more interesting than to investigate the arithmetic of such people; it would make one really understand that here there is no distinction between 20 and 21.) Could we also imagine, in contrast with the cardinal numbers a kind of number consisting of a series like the cardinal numb:rs without the 5? Certainly; but this kind of number couldn't be used for atry of the things for which we use the cardinal numbers. The way in which these numbers are missing a five is not like the way in which an apple may have been taken out of a box of apples and can be put back again; it is of their essence to lack a 5; they do not know the 5 (in the way that the cardinal numbers do not know the number t). So these numbers (if you want to call them that) would be used in cases where the cardinal numbers (with the 5) couldn't meaningfully be used. (Doesn't the nonsensicality of the talk of the "basic intuition" show itself here ?) When the intuitionists speak of the "basic intuition" - is this a psychological process? If so, how does it come into mathematics? Isn't what they mean only a primitive sign (in Frege's sense); an element of a calculus? Strange as it sounds, it is possible to know the prime numbers let's say - only up to 7 and thus to have a finite system of prime numbers. And what we call the discovery that there are infinitely many primes is in truth the discovery of a new system with no greater rights than the other. If you close your eyes and see countless glimmering spots of light coming and going, as we might say, it doesn't make sense to speak of a 'number' of simultaneously seen dots. And you can't say "there is always a definite number of spots of light there, we

just don't know what it is"; that would correspond to a rule applied in a case where you can speak of checking the number. (It makes sense to say: I divide many among many. But the proposition "I couldn't divide the many nuts among the many people" can't mean that it was logically impossible. Also you can't say "in some cases it is possible to divide many among many and in others not" ; for in that case I ask: in which cases is this possible and in which impossible? And to that no further answer can be given in the many-system.) To say of a part of my visual field that it has no colour is nonsense; and of course it is equally nonsense to say that it has colour (or a colour). On the other hand it makes sense to say it has only one colour (is monochrome, or uniform in colour) or that it has at least two colours, only two colours, etc. So in the sentence "this square in my visual field has at least two colours" I cannot substitute "one" for "two". Or again: "the square has only one colour" does not mean - on the analogy of (3x). q>x. ~ (3x, y) . q>x. cpy - "the square has one colour but not two colours". I am speaking here of the case in which it is senseless to say "that part of space has no colour". If I am counting the uniformly coloured (monochrome) patches in the square, it does incidentally make sense to say that there aren't any there at all, if the colour of the square is continually changing. In that case of course it also makes sense to say that there are one or more uniformly coloured patches in the square and also that the square has one colour and not two. - But for the moment I am disregarding that use of the sentence "the square has no colour" and am speaking of a system in which it would be called a matter of course that an area of a surface had a colour, a system, therefore, in which strictly speaking there is no such proposition. If you call the proposition self-evident you really mean something that is expressed by a grammatical rule giving the form of propositions about visual space, for

I

I

I

•

I

•

I

·

.

I

instance. If you now begin the series of statements giving the numbe~ of colours in the square with the proposition "there is one c~~our 1n the square", then of course that mustn't be the proposlt10n of grammar about the "colouredness" of space. What do you mean if you say "space is coloured"? (And aver . . ' y 1nterestmg question: what kind of question is this?) Well, perhaps you look around for confirmation and look at the different colours around you and feel the inclination to say: "wherever I look there is a .colo~r:', or "it's all coloured, all as it were painted." Here you are 1magmmg colours in contrast to a kind of colourlessness which on closer inspection turns into a colour itself. Incidentall;, when you look around for confirmation you look first and foremost at static and monochromatic parts of space, rather than at unstable unclearly coloured parts (flowing water, shadows, etc.) If you then have to admit that you call just everything that you see colour, what you want to say is that being coloured is a property of space in 1tself, not of t~~ parts of space. But that comes to the same as saying of chess that 1t 1S chess; and at best it can't amount to more than a descrip~i~n of the game. So what we must do is describe spatial ?roposltlons; but.we ca~'t justify them, as if we had to bring them mto agreement w1th an mdependent reality. In order to confirm the proposition "the visual field is coloured" one looks round and says "that there is black, and black is a colour; that is white, and white is a colour", etc. And one regards ::black is ~ colour" as like "iron is a metal" (or perhaps better, gypsum 1S a sulphur compound"). If I make it senseless to say that a part of the visual field has a colour, then asking for the analysis of a statement assigning the nu~ber of colours in a part of the visual space becomes very like askmg for the analysis of a statement of the number of parts of a rectangle that I divide up into parts by lines. Here too I can regard it as senseless to say that the rectangle "consists of no parts". Hence, one cannot say that it consists of one .or more parts, or that it has at least one part. Imagine the spec1al case of a rectangle divided by parallel lines. It doesn't matter that this is very special case, since we don't regard a game as less remarkable just because it has only a very limited applica-

don. Here I can if I want count the parts in the usual manner, and then it is meaningless to say there are parts. But I could also imagine a way of counting which so to say regards the first part as a matter of course 2 o I and doesn't count it or counts it as 0, and i j I counts only the parts which are added to this by division. Again, one could imagine I a custom according to which, say, soldiers in rank and file were always counted by o giving the number of soldiers in a line over and above the first soldier (perhaps because we wanted the number of possible combinations of the fugleman with another soldier of the rank.). But a custom might also exist of always giving the number of soldiers as I greater than the real one. Perhaps this happened originally in order to deceive a particular officer about the real number, and later came into general use as a way of counting soldiers. (The academic quarter).l The number of different colours on a surface might also be given by the number of their possible combinations in pairs and in that case the only numbers that would count would be numbers of the form ~(n - I); it would be as senseless then to talk of the 2 or 4 colours of a surface as it now is to talk of the V2 or i colours. I want to say that it is not the case that the cardinal numbers are essentially primary and what we might call the combination numbers - 1,2,6,10 etc. - are secondary. We might construct an arithmetic of the combination numbers and it would be as self-contained as the arithmetic of the cardinal numbers. But equally of course there might be an arithmetic of the even numbers or of the numbers I, 3, 4, 5, 6, 7 ... Of course the decimal system is ill-adapted for the writing of these kinds of number.

°

Imagine a calculating machine that calculates not with beads but with colours on a strip of paper. Just as we now use our fingers, or the beads on an abacus, to count the colours on a strip

a

!

324

I. This is an allusion to the German academic custom of announcing a lecture for, say, 11.15 by scheduling it "II .00 c.t." (Trs.)

so then we would use the colours on a strip to count the beads on a bar or the fingers on our hand. But how would this colour-calculating machine have to be made in order to work? We would need a sign for there being no bead on the bar. We must imagine the abacus as a practical tool and as an instrument in language. Just as we can now represent a number like 5 by the five fingers of a hand (imagine a gesture language) so we would then represent it by a strip with five colours. But I need a sign for the 0, otherwise I do not have the necessary multiplicity. Well, I can either stipulate that a black surface is to denote the 0 (this is of course arbitrary and a monochromatic red surface would do just as well): or that any one-coloured surface is to denote zero, a two-coloured surface I, etc. It is immaterial which method of denotation I choose. Here we see how the multiplicity of the beads is projected onto the multiplicity of the colours on a surface. It makes no sense to speak of a black two-sided figure in a white circl~; this is analogous to its being senseless to say that the rectangle cons1sts of 0 parts (no part). Here we have something like a lower limit of counting before we reach the number one.

'--_ _....Jlo

'-----II 11--_-1 '--L--II I . I

[]

'--_---ll

o

I

rn

'--_--II 2. '--'-......L-JII·I·I·I 1·1·1·1 I· . ·13 1 1 1 " . 1·1 I

II

I

2.

·1 ·11 . 1. I . I . IIL..-·_---J·I 4 III

IV

V

Is counting parts in I the same as counting points in IV? What makes the difference? We may regard counting the parts in I as counting rectangles; but in that case one can also say: "in this row there is no rectangle"; and then one isn't counting parts. Weare disturbed both by the analogy between counting the points and counting the parts, and by the breakdown of the analogy. There is something odd in counting the undivided surface as "one"; on the other hand we find no difficulty in seeing the surface after a single division as a picture of 2. Here we would much prefer to count "0, 2, 3", etc. And this corresponds to the seri~s of propositions "the rectangle is undivided", "the rectangle 1S divided into 2 parts", etc.

If it's a question of different colours, you can imagine a way of thinking in which you don't say that here we have two colours, but that here we have a distinction between colours; a style of thought which does not see 3 at all in red, green and yellow; which does indeed recognise as a series a series like: red; blue, green; yellow, black, white; etc., but doesn't connect it with the series I; II; III; etc., or not in such a way as to correlate I with the term red. From the point of view from which it is 'odd' to count the undivided surface as one, it is also natural to count the singly divided one as two. That is what one does if one regards it as two rectangles, and that would mean looking at it from the standpoint from which the undivided one might well be counted as one rectangle. But if one regards the first rectangle in I as the undivided surface, then the second appears as a whole with one division (one distinction) and division here does not necessarily mean dividing line. What I am paying attention to is the distinctions, and here there is a series of an increasing number of distinctions. In that case I will count the rectangles in I "0, I, 2, etc." This is all right where the colours on a strip border on each other, as in the schema I red I green I white I

But it is different if the arrangement is

~ g

g

W

W

r

r

or I wig 1 w 1 rig I r I· Of course I might also correlate each of these two schemata with the schema I wig 1 r 1

But perhaps we might refrain altogether from using a number to denote the distinction and keep solely to the schemata A, AB, ABC, etc. ; or we might describe it like this: I, 12, 123 etc., or, what comes to the same, 0, 01, 012 etc. We may very well call these too numerals. The schemataA,AB, ABC etc.,

I, II, III, etc.;

D, [[],

[II] , [ill] , etc.; 0, 1,2,3, etc.; 1,2,3, etc.; I, 12, 121323, etc., and correlate schemata like

with

the schema I wig 1 r 1 bi 1 ,etc. And that way of thinking though certainly unnatural is perfectly correct. The most natural thing is to conceive the series of schemata as A A B ABC ABC D etc. And here we may denote the first schema by '0', the second by '~', but the third say with '3', if we think of all possible distinctIOns, and the fourth by '6' Or we may call the third schema '2' (if we are concerned simply with an arrangement) and the fourth' 3'. . V!e c~ describe the way a rectangle is divided by saying: it is dIVIded Into five parts, or: 4 parts have been cut off it, or: its division-schema is ABCDE, or: you can reach every part by crossing four boundaries or: the rectangle is divided (i.e. into 2 parts), one part is divided again, and both parts of this part divided, etc. I want to show that there isn't only one method of describing the way it is divided.

etc., are all equally fundamental. We are surprised that the number-schema by which we count soldiers in a barracks isn't supposed also to hold for the parts of a rectangle. But the schema for the soldiers in the barracks is

D , D , D , etc., the one for the parts of the rectangle is r - I r - r l ITO etc. Neither is primary in comparison L-J,LU, , with the other. I can compare the series of division-schemata with the series I, 2, 3, etc. as well as with the series 0, I, 2, 3> etc. If I count the parts, then there is no ° in my number series because the series A

A B ABC etc. begins with one letter whereas the series

0,0 , 0L.:..J ,

etc. does not begin with one dot. On the other hand, I can represent any fact about the division by this series too, only in that case "I'm not counting the parts". A way of expressing the problem which, though incorrect, is natural is: why can one say "there are 2 colours on this surface" but not "there is one colour on this surface?" Or: how must I express the grammatical rule so that it is obvious and so that I'm not any longer tempted to talk nonsense? Where is the false thought, the

329

I.

false analogy by which I am misled into misusing language? How must I set out the grammar so that this temptation ceases? I think that setting it out by means of the series

0

A AB ABC and so on

and

D

B

and so on removes the unclarity. What matters is whether in order to count I use a number series that begins with 0 or one that begins with 1. It is the same if I am counting the lengths of sticks or the size of hats. H I counted with strokes, I might write them thus

I, V ,

\V,

~

, in order to show that what matters is the distinction

between the directions and that a simple stroke corresponds to o (i.e. is the beginning). Here incidentally there is a certain difficulty about the numerals + 1), etc.: beyond a certain length we cannot distinguish them any further without counting the strokes, and so without translating the signs into different ones. "1111111 j II" and" 11111111111" cannot be distinguished in the same sense as 10 and 11, and so they aren't in the same sense distinct signs. The same thing could also happen incidentally in the decimal system (think of the numbers 1111111111 and I I I I I I 1 I I I 1), and that is not without significance.

(1), ((1)

Imagine someone giving us a sum to do in a stroke-notation, say + 11111111111, and, while we are calculating, amusing himself by removing and adding strokes without our noticing. He would keep on saying: "but the sum isn't right", and we would keep going through it again, fooled every time. - Indeed, strictly speaking, we wouldn't have any concept of a criterion for the correctness of the calculation. 1111111111

Here one might raise questions like: is it only very probable that 464 + Z7Z = 736? And in that case isn't z + 3 = 5 also only very probable? And where is the objective truth which this probability approaches? That is, how do we get a concept of z + 3's really being a certain number, apart from what it seems to us to be? For if it were asked: what is the criterion in the stroke-notation for our having the same numeral in front of us twice? - the answer might be: "if it looks the same both times" or "if it contains the same number of lines both times". Or should it be: if a one-one correlation etc. is possible? How can I know that 1111111111 and 1111111111 are the same sign? After all it is not enough that they look alike. For having roughly the same gestalt can't be what is to constitute the identity of the signs, but just their being the same in number. (The problem of the distinction between 1 + 1 + 1 + I + 1 + 1 + I and I + I + I + I + I + I + I + I is much more fundamental than appears at first sight. It is a matter of the distinction between physical and visual number.)

A cardinal number is an internal property of a list. Are numbers essentially concerned with concepts? I believe this amounts to asking whether it makes sense to ascribe a number to objects that haven't been brought under a concept. Does it, for example, make sense to say "a, band c are three objects"? - Admittedly we have a feeling: why talk about concepts, the number of course depends only on the extension of the concept, and once that has been determined the concept may drop out of the picture. The concept is only a method for determining an extension, but the extension is autonomous and, in its essence, independent of the concept; for it's quite immaterial which concept we have used to determine the extension. That is the argument for the extensional viewpoint. The immediate objection to it is: if a concept is really only an expedient for aiming at an extension, then there is no place for concepts in arithmetic; in that case we must simply divorce a class completely from the concept which happens to be associated with it. But if it isn't like that, then an extension independent of the concept is just a chimaera, and in that case it's better not to speak of it at all, but only of the concept. The sign for the extension of a concept is a list. We might say, as an approximation, that a number is an external property of a concept and an internal property of its extension (the list of objects that fall under it). A number is a schema for the extension of a concept. That is, as Frege said, a statement of number is a statement about a concept (a predicate). It's not about the extension of a concept, i.e. a list that may be something like the extension of a concept. But a number-statement about a concept has a similarity to a proposition saying that a determinate list is the extension of the concept. I use such a list when I say "a, b, c, d, fall under the concept F(x)": "a, b, c, d," is the list. Of course this proposition

says the same as Fa.Fb.Fc.Fd; but the use of the list in writing the proposition shows its relationship to "(3x, y, z, u). Fx. Fy. Fz. Fu" which we can abbreviate as "(3111Ix). F(x)." What arithmetic is concerned with is the schema 1111. - But does arithmetic talk about the lines that I draw with pencil on paper? - Arithmetic doesn't talk about the lines, it operates with them. A statement of number doesn't always contain a generalization or indeterminacy: "The line AB is divided into 2 (3, 4, etc.) equal parts." If you want to know what 2 + 2 = 4 means, you have to ask how we work it out. That means that we consider the process of calculation as the essential thing; and that's how we look at the matter in ordinary life, at least as far as concerns the numbers that we have to work out. We mustn't feel ashamed of regarding numbers and sums in the same way as the everyday arithmetic of every trader. In everyday life we don't work out 2 + 2 = 4 or any of the rules of the multiplication table; we take them for granted like axioms and use them to calculate. But of course we could work out 2 + 2 = 4 and children in fact do so by counting off. Given the sequence of numbers I 2 3 4 5 the calculation is I 2 I 2 12 3 4 Abbreviative Definitions: (3x). cpx: _ (3x, y). cpx. cpy. Def .( ex). cpx (3x, y). cpx. cpy: - (3x, y,z). cpx. cpy. cpz. Def .(ex, y). cpx. cpy, (ex). cpx. ~ .(e;lx). cpx (ex,y).cpx.cpy. Def .(e;llx).cpx.

Def

.(e;2x).cpx,

etc.

etc.

It can be shewn that (ellx). cpx.( ell Ix) .ljJx. - (3x). cpx .ljJx. :::> .( elllllx). cpx V IjJx '-v--'

Ind. is a tautology.

332

333

Does that prove the arithmetical proposition 2 + 3 = 5? Of course not. It does not even show that (ellx). rpx.(elllx). ~x. Ind .. :::J .(ell + Illx). rpx v ~x is tautologous, because nothing was said in our definitions about a sum (II + III). (I will write the tautology in the abbreviated form "ell. eili. :::J. ell III"·) Suppose the question is, given a left hand side, to find what number of lines to the right of" :::J" makes the whole a tautology. We can find the number, we can indeed discover that in the case above it is II + III; but we can equally well discover that it is 1+ IIII or I + III + I, for it is all of these. We can also find an inductive proof that the algebraic expression en.em.:::J .en+m is tautologous. Then I have a right to regard a proposition like eI7.e28. :::J . e(17 + 28) as a tautology. But does that give us the equation 17 + 28 = 45? Certainly not. I still have to work it out. In accordance with this general rule, it also makes sense to write e2. e3 . :::J . e5 as a tautology if, as it were, I don't yet know what 2 + 3 yields; for 2 + 3 only has sense in so far as it has still to be worked out. Hence the equation II + III = IIIII only has a point if the sign "11111" can be recognised in the same way as the sign" 5", that is, independently of the equation. The difference between my point of view and that of contemporary writers on the foundations of arithmetic is that I am not obliged to despise particular calculi like the decimal system. For me one calculus is as good as another. To look down on a particular calculus is like wanting to play chess without real pieces, because playing with pieces is too particularized and not abstract enough. If the pieces really don't matter then one lot is just as good as another. And if the games are really distinct from each other, then one game is as good, i.e. as interesting, as the other. None of them is more sublime than any other.

334

Which proof of ell. eili. :::J. ellill expresses our knowledge that this is a correct logical proposition? Obviously, one that makes use of the fact that one can treat (3x) ... as a logical sum. We may translate from a symbolism like

~

("if there is a star in each square, then there are two

in the whole rectangle") into the Russellian one. And it isn't as if the tautologies in that notation expressed an idea that is confirmed by the proof after first of all appearing merely plausible; what appears plausible to us is that this expression is a tautology (a law of logic). The series of propositions (3x): aRx. xRb (3x, y): aRx.xRy. yRb (3x,y,z):aRx.xRy.yRz.zRb, etc. may perfectly well be expressed as follows: "There is one term between a and b". "There are two terms between a and b", etc., and may be written in some such way as: (3 IX) . aRxRb, (3 2X) . aRxRb, etc. But it is clear that in order to understand this expression we need the explanation above, because otherwise by analogy with (32X).rpX.= .(3x, y).rpx.rpy you might believe that (32x).aRxRb was equivalent to the expression (3x, y). aRxRb. aRyRb. Of course I might also write "(32X, y). F(x, y)" instead of "(3x, y) .F(x, y)". But then the question would be: what amI to take "(33X, y).F(x, y)" as meaning? But here a rule can be given; and indeed we need one that takes us further in the number series as far as we want to go. E.g.: (3~x, y).F(x, y). = .(3x, y, z):F(x, y).F(x, z).F(y, z) (34x,y).F(x,y). = .(3x,y,z,u):F(x,y).F(x,z) .... followed by the combinations of two elements, and so on. But we might also give the following definition: (33x,y).F(x,y). = .(3x,y,z):F(x,y).F(y,x).F(x,z). F(z,x).F(y,z).F(x,y), and so on.

335

"(33 x, y).F(x, y)" would perhaps correspond to the proposition in word-language "F(x, y) is satisfied by 3 things"; and that proposition too would need an explanation if it was not to be ambiguous. Am I now to say that in these different cases the sign "3" has different meanings? Isn't it rather that the sign" 3" expresses whatis common to the different interpretations? Why else would I have chosen it? Certainly, in each of these contexts, the same rules hold for the sign" 3". It is replaceable by 2 + I as usual and so on. But at all events a proposition on the pattern of ell. eili. ::> • ellill is no longer a tautology. Two men who live at peace with each other and three other men who live at peace with each other do not make five men who live at peace with each other. But that does not mean that 2 + 3 are no longer 5; it is just that addition cannot be applied in that way. For one might say: 2 men who ... and 3 men who ... , each of whom lives at peace with each of the first group, = 5 men who ... In other words, the signs of the form (3 IX, y). F(x, y), (32X, y). F(x, y) etc. have the same multiplicity as the cardinal numbers, like the signs (3IX).N". (We might call the second and third lines of the rule R taken together the operation, like the second and third term of the sign N.) Thus too the explanation of the use of the recursive definition "a + (b + 1) = (a + b) + 1" is a part of that rule itself; or if you like a repetition of the rule in another form; just as "1, I + I, I + I + I and so on" means the same as (i.e. is translatable into) "I I, ;, ; + 1I". The translation into word-language casts light on the calculus with the new signs, because we have already mastered the calculus with the signs of word-language. The sign of a rule, like any other sign, is a sign belonging to a calculus; its job isn't to hypnotize people into accepting an application, but to be used in the calculus in accordance with a system. Hence the exterior form is no mor.e essential than that of an arrow --+; what is essential is the system in which the sign for the rule is employed. The system of contraries - so to speak - from which the sign is distinguished etc. What I am here calling the description of the application is itself of course something that contains an "and so on", and so it can itself be no more than a supplement to or substitute for the rulesign. What is the contradictory of a general proposition like a + (b f(I + I)) = a + «b + I) + I)? What is the system of propositions within which this proposition is negated? Or again, how, and in what form, can this proposition come into contradiction with others? What question does it answer? Certainly not the question I.

432

I

{,

Cf. footnote, p.

420.

433

whether (n). fn or (3n). - fn is the case, because it is the rule R that contributes to the generality of the proposition. The generality of a rule is eo ipso incapable of being brought into question. Now imagine the general rule written as a series

PII' PI2' PI3 .. . P21, P22' P23 .. . P3l' P32' P33 .. . and then negated. If we regard it as (x). fx, then we are treating it as a logical product and its opposite is the logical sum of the denials ofplI' PI2 etc. This disjunction can be combined with any random product PII' P21' P22 ... Pmn' (Certainly if you compare the proposition with a logical product, it becomes infinitely significant and its opposite void of significance). (But remember that the "and so on" in the proposition comes after a comma, not after an "and" (" .") The "and so on" is not a sign of incompleteness.) Is the rule R infinitely significant? Like an enormously long logical product? That one can run the number series though the rule is a form that is given; nothing is affirmed about it and nothing can be denied about it. Running the stream of numbers through is not something which I can say I can prove. I can only prove something about the form, or pattern, through which I run the numbers. But can't we say that the general number rule a + (b + c) = (a+ b) + c ... A) has the same generality as a+(1 + 1) =(a + 1) + 1 (in that the latter holds for every cardinal number and the former for every triple of cardinal numbers) and that the inductive proof of Ajustiftes the rule A? Can we say that we can give the rule A, since the proof shows that it is always right? Does l/3 = 0'3 .!. justify the rule 113 = 0'3, 1/3 = 0'333, 1J3 = 0.333 and so on?" ... P) A is a completely intelligible rule; just like the replacement rule P. But I can't give such a rule, for the reason that I can already cal-

434

culate the particular instances of A by another rule; just as I cannot give P as a rule if! have given a rule whereby I can calculate 1 3 = o· 3 etc.

t

How would it be if someone wanted to lay down" 2 5 x 25 = 625" as a rule in addition to the multiplication rules. (I don't say "25 x 25 = 624"!) - 25 x 25 = 625 only makes sense if the kind of calculation to which the equation belongs is already known, and it only makes sense in connection with that calculation. A only makes sense in connection with A's own kind of calculation. For the first question here would be: is that a stipulation, or a derived proposition? If 25 x 25 = 625 is a stipulation, then the multiplication sign does not mean the same as it does, e.g. in reality (that is, we are dealing with a different kind of calculation). And if A is a stipulation, it doesn't define addition in the same way as if it is a derived proposition. For in that case the stipulation is of course a definition of the addition sign, and the rules of calculation that allow A to be worked out are a different definition of the same sign. Here I mustn't forget that IX, ~, Yisn't the proof of A, but only the form of the proof, or of what is proved; so IX, ~, Yis a definition ofA. Hence I can only say" 2 5 x 25 = 625 is proved" if the method of proof is fixed independently of the specific proof. For it is this method that settles the meaning of "~ x Y)" and so settles 1vhat is proved. So to that extent the form.!!. b = c belongs to the method ~

of proof that explains the sense of c. Whether I have calculated correctly is another question. And similarly IX, ~, Y belong to the method of proof that defines the sense of the proposition A. Arithmetic is complete without a rule like A; without it it doesn't lack anything. The proposition A is introduced into arithmetic with the discovery of a periodicity, with the construction of a new calculus. Before this discovery or construction a question about the correctness of that proposition would have as little sense as a question about the correctness of "1/3 = o· 3, 1/3 = 0·33 ... ad inf."

435

The stipulation of P is not the same thing as the proposition "1/3 = o· 3" and in that sense "a + (b + c) = (a + b) + c) is different from a rule (stipulation) such as A. The two belong to different calculi. The proof of 0(, ~, Y is a proof or justification of a rule like A only in so far as it is the general form of the proof of arithmetical propositions of the form A. Periodicity is not a sign (symptom) of a decimal's recurring; the expression "it goes on like that for ever" is only a translation of the sign for periodicity into another form of expression. (If there was something other than the periodic sign of which periodicity was only a symptom, that something would have to have a specific expression, which could be nothing less than the complete expression of that something.)

37

Seeing or viewing a sign in a particular manner. Discovering an aspect of a mathematical expression. "Seeing an expression in a particular wcry". Marks of emphasis. Earlier I spoke of the use of connection lines, underlining etc. to bring out the corresponding, homologous, parts of the equations of a recursion proof. In the proof Y

IX

a + (b + I) = (a + b) + I

~8

I

!='

I

a+(b+(c+ 1))=(a+(b+c))+ ~

(a+ b)+(.:....~j)=((a+ b)+ ~)+

~

I

I
p. q constructs a new sign, the sign for that rule. (I am assuming that a calculus with p, q, :::>, has already been in use, and that this rule is now added to make it a new calculus.) It is true that the notation "x2" takes away the possibility of replacing one of the factors x by another number. Indeed, we could imagine two stages in the discovery (or construction) of x2 • At first, people might have written "x=" instead of "x2", before it occurred to them that there was a system x. x, x. x . x, etc.; later, they might have hit upon that too. Similar things have occurred in mathematics countless times. (In Liebig's sign for an oxide oxygen did not appear as an element in the same way as what was oxidized. Odd as it sounds, we might even today, with all the data available to us, give oxygen a similarly privileged position only, of course, in the form of representation - by adopting an incredibly artificial interpretation, i.e. grammatical construction.) The definitions x. x = X2, X. x. X = x 3 don't bring anything into the world except the signs "x2" and "x 3 " (and thus so far it isn't necessary to write numbers as exponents). I

Of course L\. can be regarded as a definition! Because the sign on the left hand side is in fact used, and there's no reason why we shouldn't abbreviate it according to this convention. Only in that case either the sign on the right or the sign on the left is used in a way different from the one now usual. It can never be sufficiently emphasized that totallY different kinds of sign-rules get written in the form of an equation.

The 'definition' x. x = x2 might be regarded as merely allowing us to replace the sign "x. x" by the sign "x2 , " like the definition " I + I = 2"; but it can also be regarded (and in fact is regarded) as allowing us to put a2 instead of a.a, and (a + b)2 instead of (a+ b).(a+ b) and in such a way that any arbitrary number can be substituted for the x.

The process of generalization creates a new sign-system. 1

Of course Sheffer's discovery is not the discovery of the definition ~ p.~ q = plq. Russell might well have given that definition without being in possession of Sheffer's system, and on the other hand Sheffer might have built up his system without the definition. His system is contained in the use of the signs "~p.~p" for "~p" and "~(~p .~q) .~(~p .~q)"for"p v q;'andall "plq" doesis to permit an abbreviation. Indeed, we can say that someone could well have been acquainted with the use of the sign "~(~p.~q). ~(~p. ~q)" for "p v q" without recognizing the system pi q ·1· p 1q in it.

443

It makes matters clearer if we adopt Frege's two primitive signs "~" and" . ". The discovery isn't lost if the definitions are written ~p .~p = ~p and ~(~p.~p) .~( ~q .~q) = p. q. Here apparently nothing at all has been altered in the original signs. But we might also imagine someone's having written the whole Fregean or Russellian logic in this system, and yet, like Frege, calling "~,, and "." his primitive signs, because he did not see the other system in his proposition. It is clear that the discovery of Sheffer's system in ~. p. ~p = ~p and ~(~p. ~p) . ~(~q . ~q) = p. q corresponds to the discovery that x 2 + ax + a2 is a specific instance of a 2 + zab + b 2 • 4

We don't see that something can be looked at in a certain way until it is so looked at. We don't see that an aspect is possible until it is there.

Moreover, if I say that I understand the equations as particular cases of the rule, my understanding has to be the understanding that shows itself in the explanations of the relations between the rule and the equations, i.e. what we express by the substitutions. If I don't regard that as an expression of what I understand, then nothing is an expression of it; but in that case it makes no sense either to speak of understanding or to say that I understand something definite. For it only makes sense to speak of understanding in cases where we understand one thing as opposed to another. And it is this contrast that signs express. Indeed, seeing the internal relation must in its turn be seeing something that can be described, something of which one can say: "I see that such and such is the case"; it has to be really something of the same kind as the correlation-signs (like connecting lines, brackets, substitutions, etc.). Everything else has to be contained in the application of the sign of the general rule in a particular case.

That sounds as if Sheffer's discovery wasn't capable of being represented in signs at all. (Periodic division.) But that is because we can't smuggle the use of the sign into its introduction (the rule is and remains a sign, separated from its application).

It is as if we had a number of material objects and discovered they had surfaces which enabled them to be placed in a continuous row. Or rather, as if we discovered that such and such surfaces, which we had seen before, enabled them to be placed in a continuous row. That is the way many games and puzzles are solved.

Of course I can only apply the general rule for the induction proof when I discover the substitution that makes it applicable. So it would be possible for someone to see the equations

The person who discovers periodicity invents a new calculus. The question is, how does the calculus with periodic division differ from the calculus in which periodicity is unknown?

(a + I) + 1 = (a + I)

1 + (a + I) = (I

444

+1 + a) + 1

(We might have operated a calculus with cubes without having had the idea of putting them together to make prisms.)

445

(On: The process of generalization creates a new sign-system)

regard A as the result of a transformation of those equations. (Whereas of course in reality I regard the signs IX, ~, Y in quite a different way, which means that the c in {3 and y isn't used as a variable in the same way as a and b. Hence one can express this new view ofB, by saying that the c does not occur in A.)

It is a very important observation that the c in A is not the same variable as the c in ~ and y. So the way I wrote out the proof was not quite correct in a respect which is very important for us. In A we could substitute n for c, whereas the cs in ~ and yare identical. But another question arises: can I derive from A that i + (k + c) = (i + k) + c? If so, why can't I derive it in the same way from B? Does that mean that a and b in A are not identical with a and b in IX, ~ and y?

What I said about the new way of regarding IX, ~, Y might be put like this: IX is used to build up ~ and y in exactly the same way as the fundamental algebraic equations are used to build up an equation like (a + by = a 2 + 2ab + b 2 • But if that is the way they are derived, we are regarding the complex IX ~ Y in a new way when we give the variable c a function which differs from that of a and b (c becomes the hole through which the stream of numbers has to flow).

Appendix l

We see clearly that the variable c in B isn't identical with the c in A if we put a number instead of it. Then B is something like

4+(5+1) =(4+5)+1 ) 4 + (5 + (6 + I)) = (4 + (5 + 6)) + 1 Y (4 + 5) + (6 + I) = «4 + 5) + 6) + 1 IX

~

... W

but that doesn't have corresponding to it an equation like Aw: 4+(5 + 6) =(4+ 5) + 6! What makes the induction proof different from a proof of A is expressed in the fact that the c in B is not identical with the one in A, so that we could use different letters in the two places. All that is meant by what I've written above is that the reason it looks like an algebraic proof of A is that we think we meet the same variables a, b, c in the equations A as in IX, ~, Y and so we

I. Remarks taken from the Manuscript volume. We must not forget that Wittgenstein omitted them. Even in the MS they are not set out together as they are here. (Ed.)

447

38

Proof fry induction, arithmetic and algebra

Why do we need the commutative law? Not so as to be able to write the equation 4 + 6 = 6 + 4, because that equation is justified by its own particular proof. Certainly the proof of the commutative law can also be used to prove it, but in that case it becomes just a particular arithmetical proof. So the reason I need the law, is to apply it when using letters. And it is this justification that the inductive proof cannot give me. However, one thing is clear: if the recursive proof gives us the right to calculate algebraically, then so does the arithmetical proof LI. Again: the recursive proof is - of course - essentially concerned with numbers. But what use are numbers to me when I want to operate purely algebraically? Or again, the recursion proof is only of use to me when I want to use it to justify a step in a numbercalculation. But someone might ask: do we need both the inductive proof and the associative law, since the latter cannot provide a foundation for calculation with numbers, and the former cannot provide one for transformations in algebra? Well, before Skolem's proof was the associative law, for example, just accepted without anyone's being able to work out the corresponding step in a numerical calculation? That is, were we previously unable to work out 5 + (4 + 3) = (5 + 4) + 3, and did we treat it as an axiom?

like if it had been assumed as an axiom instead of being proved? What would a proposition be like that permitted one to put 5 + (7 + 9) = (5 + 7) + 9) without being able t~ 'prove it? It is obvious that there never has been such a proposltlon. But couldn't we also say that the associative law isn't used at all in arithmetic and that we work only with particular number calculations? Even when algebra uses arithmetical notation, it is a totally different calculus, and cannot be derived from the arithmetical one. To the question "is 5 X 4 = 2O"? one might answer: "let's check whether it is in accord with the basic rules of arithmetic" and similarly I might say: let's check whether A is in accord with the basic rules. But with which rules? Presumably with r:1.. But before we can bring (J. and A together we need to stipulate what we want to call "agreement" here. That means that (J. and A are separated by the gulf between arithmetic and algebra,l and if B is to count as a proof of A, this gulf has to be bridged over by a stipulation. It is quite clear that we do use an idea of this kin~ of agreement when for instance, we quickly work out a numencal example to check the correctness of an algebraic proposition. And in this sense I might e.g. calculate 25XI6 I6x25 25 32 150

80

400

4 00

If I say that the periodic calculation proves the proposition that justifies me in those steps, what would the proposition have been I.

I.

Above, p. 43 I.

To repeat,

ot is: a + (b + I) = (a + b) + I A is: a + (b + c) = (a + b) + c. (Ed.)

449

and say: "yes, it's right, a. b is equal to b. a" - if I imagine that I have forgotten. Considered as a rule for algebraic calculation, A cannot be proved recursively. We would see that especially clearly if we wrote down the "recursive proof" as a series of arithmetical expressions. Imagine them written down (i.e. a fragment of the series plus "and so on") without any intention of "proving" anything, and then suppose someone asks: "does that prove a + (b + c) = (a + b) + c ?". We would ask in astonishment "How can it prove anything of the kind? The series contains only numbers, it doesn't contain any letters". - But no doubt we might say: if I introduce A as a rule for calculation with letters, that brings this calculus in a certain sense into unison with the calculus of the cardinal numbers, the calculus I established by the law for the rules of addition (the re n' the rules for checking a proposition of that kind - e.g. '32 > Ii. If I say: "given any n there is a ~ for which the function is less than n", I am ipso facto referring to a general arithmetical criterion that indicates when F(~) < n. If in the nature of the case I cannot write down a number independently of a number system, that must be reflected in the general treatment of number. A number system is not something inferior -like a Russian abacus - that is only of interest to elementary schools while a more lofty general discussion can afford to disregard it. Again, I don't lose anything of the generality of my account if I give the rules that determine the correctness and incorrectness (and thus the sense) of 'm > n' for a particular system like the decimal system. After all I need a system, and the generality is preserved by giving the rules according to which one system can be translated into another. A proof in mathematics is general if it is generally applicable. You can't demand some other kind of generality in the name of rigour. Every proof rests on particular signs, produced on a particular occasion. All that can happen is that one type of generality may appear more elegant than another. ((Cf. the employment of the decimal system in proofs concerning ~ and 11». "Rigorous" means: clear.!

I. (Remark in the margin in pencil.) A defence, against Hardy, of the decimal system in proofs, etc.

"We may imagine a mathematical propo~ition as a cre~ture whic~ itself knows whether it is true or false (m contrast wIth propoSItions of experience). . . . A mathematical proposition itself knows that 1t IS true or that It is false. Ifitis about all numbers, it must also survey all the numbers. "Its truth or falsity must be contained in it as is its sense." "It's as though the generality of a proposition like '(n). E:~n)' were only a pointer to the genuine, actual, mathematical general~ty, and not the generality itself. As if the proposition fo~med a s~gn only in a purely external way and you still needed to glVe the sIgn a sense from within." "We feel the generality possessed by the math~~atical asse~:ion to be different from the generality of the proposltlon proved. "We could say: a mathematical proposition is an allusion to a proof."! What would it be like if a proposition itself did not quite gras? its sense? As if it were, so to speak, too grand for itself? That IS really what logicians suppose. A proposition that deals with all numbers ca~n?t b~ thought ~f as verified by an endless striding, for, if the strldmg IS endless, It does not lead to any goal. Imagine an infinitely long row of tre~s, and, so that we can inspect them, a path beside them. All nght, the path must b,e endless. But if it is endless, then that means precisely that y~~ can t walk to the end of it. That is, it does not put me in a pOSItiOn to survey the row. That is to say, the endless path does not have an end 'infinitely far away', it has no end. Nor can you say: "A proposition cannot deal with all the numbers one by one, so it has to deal with them by means of the concept of number" as if this were a pis aller: "Beca~se. w~ can't do it like this, we have to do it another way." But It IS mdee~ possible to deal with the numbers one by one, only that doesn t I.

454

Philosophical Remarks, 122, pp. 143- 145.

455

lead to the totality. That doesn't lie on the path on which we go step by step, not even at the infinitely distant end of that path. (This all only means that "e:(o). e:(l). e:(z) and so on" is not the sign for a logical product.) "It cannot be a contingent matter that all numbers possess a property; if they do so it must be essential to them." - The proposition "men who have red noses are good-natured" does not have the same sense as the proposition "men who drink wine are goodnatured" even if the men who have red noses are the same as the men who drink wine. On the other hand, if the numbers m, n, 0 are the extension of a mathematical concept, so that is the case that fm. fn . fo, then the proposition that the numbers that satisfy f have the property e: has the same sense as "e:(m).e:(n)e:.(o)". This is because the propositions "f(m).f(n).f(o)" and "e(m). e(n). e(o)" can be transformed into each other without leaving the realm of grammar. Now consider the proposition: "all the n numbers that satisfy the condition F(~) happen by chance to have the property e:". Here what matters is whether the condition F(~) is a mathematical one. If it is, then I can indeed derive e:(x) from F(x), if only via the disjunction of the n values ofF(~). (For what we have in this case is in fact a disjunction). So I won't call this chance. - On the other hand if the condition is a non-mathematical one, we can speak of chance. For example, if! say: all the numbers I saw today on buses happened to prime numbers. (But, of course, we can't say: the numbers 17, 3, 5, 31 happen to be prime numbers" any more than "the number 3 happens to be a prime number"), "By chance" is indeed the opposite of "in accordance with a general rule", but however odd it sounds one can say that the proposition "17, 3,5,31 are prime numbers" is derivable by a general rule just like the proposition z + 3 = 5.

If we now return to the first proposition, we may ask again: How is the proposition "all numbers have the property e:" supposed to be meant? How is one supposed to be able to know? For to settle its sense you must settle that too! The expression "by chance" indicates a verification by successive tests, and that is contradicted by the fact that we are not speaking of a finite series of numbers.

In mathematics description and object are equivalent. "The fifth number of the number series has these properties" says the same as "5 has these properties". The properties of a house do not follow from its position in a row of houses; but the properties of a number are the properties of a position. You might say that the properties of a particular number cannot be foreseen. You can only see them when you've got there. What is general is the repetition of an operation. Each stage of the repetition has its own individuality. But it isn't as if I use the operation to move from one individual to another so that the operation would be the means for getting from one to the other like a vehicle stopping at every number which we can then study: no, applying the operation +1 three times yields and is the number 3· (In the calculus process and result are equivalent to each other.) But before deciding to speak of "all these individualities" or "the totality of these individualities" I had to consider carefully what stipulations I wanted to make here for the use of the expressions "all" and "totality". It is difficult to extricate yourself completely from the extensional viewpoint: You keep thinking "Yes, but there must still be an internal relation between x 3 + y3 and Z3 since at least extensions of these expressions if I only knew them would have to show the result of such a relation". Or perhaps: "It must surely be either essential to all numbers to have the property or not, even if I can't know it."

457

"If I run through the number series, I either eventually come to a number with the property E or I never do." The expression "to run through the number series" is nonsense; unless a sense is give1J to it which removes the suggested analogy with "running through the numbers from I to 100".

It is differently verified and so is of a different kind. The verification is not a mere token of the truth, but determines the sense of the proposition. (Einstein: how a magnitude is measured is what it is.)

When Brouwer attacks the application of the law of excluded middle in mathematics, he is right in so far as he is directing his attack against a process analogous to the proof of empirical propositions. In mathematics you can never prove something like this: I saw two apples lying on the table, and now there is only Dne there, so A has eaten an apple. That is, you can't by excluding certain possibilities prove a new one which isn't already contained in the exclusion because of the rules we have laid down. To that extent there are no genuine alternatives in mathematics. If mathematics was the investigation of empirically given aggregates, one could use the exclusion of a part to describe what was not excluded, and in that case the non-excluded part would not be equivalent to the exclusion of the others. The whole approach that if a proposition is valid for one region of mathematics it need not necessarily be valid for a second region as well, is quite out of place in mathematics, completely contrary to its essence. Although many authors hold just this approach to be particularly subtle and to combat prejudice. It is only if you investigate the relevant propositions and their proofs that you can recognize the nature of the generality of the propositions of mathematics that treat not of "all cardinal numbers" but e.g. of "all real numbers".

How a proposition is verified is what it says. Compare generality in arithmetic with the generality of non-arithmetical propositions.

459

40 On set theory

A misleading picture: "The rational points lie close together on the number-line." Is a space thinkable that contains all rational points, but not the irrational ones? Would this structure be too coarse for our space, since it would mean that we could only reach the irrational points approximately? Would it mean that our net was not fine enough? No. What we would lack would be the laws, not the extensions. Is a space thinkable that contains all rational points but not the irrational ones? That only means: don't the rational numbers set a precedent for the irrational numbers? No more than draughts sets a precedent for chess. There isn't any gap left open by the rational numbers that is filled up by the irrationals. We are surprised to find that "between the everywhere dense rational points", there is still room for the irrational~. (What balderdash!) What does a construction like that for V 2 show? Does it show how there is yet room for this point in between all the rational points? It shows that the point yielded by the construction, yielded by this construction, is not rational. - And what corresponds to this construction in arithmetic? A sort of number which manages ajter allto squeeze in between the rational numbers ? A law that is not a law of the nature of a rational number. The explanation of the Dedekind cut pretends to be clear when it says: there are 3 cases: either the class R has a first member and L no last member, etc. In fact two of these 3 cases cannot be imagined, unless the words "class", "first member", "last member", altogether change the everyday meanings thay are supposed to have retained.

That is, if someone is dumbfounded by our talk of a class of points that lie to the right of a given point and have no beginning, and says: give us an example of such a class - we trot out the class of rational numbers; but that isn't a class of points in the original sense. The point of intersection of two curves isn't the common member of two classes of points, it's the meeting of two laws. Unless, very misleadingly, we use the second form of expression to define the first. After all I have already said, it may sound trivial if I now say that the mistake in the set-theoretical approach consists time and again in treating laws and enumerations (lists) as essentially the same kind of thing and arranging them in parallel series so that one fills in gaps left by another. The symbol for a class is a list. Here again, the difficulty arises from the formation of mathematical pseudo-concepts. For instance, when we say that we can arrange the cardinal numbers, but not the rational numbers, in a series according to their size, we are u1?-consciously presupposing that the concept of an ordering by size does have a sense jor rational numbers, and that it turned out on investigation that the ordering was impossible (which presupposes that the attempt is thinkable).Thus one thinks that it is possible to attempt to arrange the real numbers (as if that were a concept of the same kind as 'apple on this table') in a series, and now it turned out to be impracticable. For its form of expression the calculus of sets relies as far as possible on the form of expression of the calculus of cardinal numbers. In some ways that is instructive, since it indicates certain formal similarities, but it is also misleading, like calling something a knife that has neither blade nor handle. (Lichtenberg.)

(The only point there can be to elegance in a mathematical proof is to reveal certain analogies in a particularly striking manner, when that is what is wanted; otherwise it is a product of stupidity and its only effect is to obscure what ought to be clear and manifest. The stupid pursuit of elegance is a principal cause of the mathematicians' failure to understand their own operations; or perhaps the lack of understanding and the pursuit of elegance have a common origin.) Human beings are entangled all unknowing in the net of language. "There is a point where the two curves intersect." How do you know that? If you tell me, I will know what sort of sense the proposition "there is ... " has. If you want to know what the expression "the maximum of a curve" means, ask yourself: how does one find it? - If something is found in a different way it is a different thing. We define the maximum as the point on the curve higher than all the others, and from that we get the idea that it is only our human weakness that prevents us from sifting through the points of the curve one by one and selecting the highest of them. And this leads to the idea that the highest point among a finite number of points is essentially the same as the highest point of a curve, and that we are simply finding out the same thing by two different methods, just as we find out in two different ways that there is no one in the next room; one way if the door is shut and we aren't strong enough to open it, and another way if we can get inside. But, as I said, it isn't human weakness that's in question where the alleged description of the action "that we cannot perform" is senseless. Of course it does no harm, indeed it's very interesting, to see the analogy between the maximum of a curve and the maximum (in another sense) of a class of points, provided that the analogy doesn't instil the prejudice that in each case we have fundamentally the same thing.

It's the same defect in our syntax which presents the geometrical proposition "a length may be divided by a point into two parts" as a proposition of the same form as "a length may be divided for ever"; so that it looks as if in both cases we can say "Let's suppose the possible division to have been carried out". "Divisible into two parts" and "infinitely divisible" have quite different grammars. We mistakenly treat the word "infinite" as if it were a number word, because in everyday speech both are given as answers to the questron "how many?" "But after all the maximum is higher than any other arbitrary points of the curve." But the curve is not composed of points, it is a law that points obey, pr again, a law according to which points can be constructed. If you now ask: "which points?" I can only say, "well, for instance, the points P, Q, R, etc." On the one hand we can't give a number of points and say that they are all the points that lie on the curve, and on the other hand we can't speak of a totality of points as something describable which although we humans cannot count them might be called the totality of all the points of the curve - a totality too big for us human beings. On the one hand there is a law, and on the other points on the curve; - but not "all the points of the curve". The maximum is higher than any point of the curve that happens to be constructed, but it isn't higher than a totality of points, unless the criterion for that, and thus the sense of the assertion, is once again simply construction according to the law of the curve. Of course the web of errors in this region is a very complicated one. There is also e.g. the confusion between two different meanings of the word "kind". We admit, that is, that the infinite numbers are a different kind of number from the finite ones, but then we misunderstand what the difference between different kinds amounts to in this case. We don't realise, that is, that it's not a matter of distinguishing between objects by their properties in the way we distinguish between red and yellow apples, but a matter of different logical forms. - Thus Dedekind tried to describe an infinite class

by saying that it is a class which is similar to a proper subclass of itself. Here it looks as if he has given a property that a class must have in order to fall under the concept "infinite class" (Frege).1 Now let us consider how this definition is applied. I am to investigate in a particular case whether a class is finite or not, whether a certain row of trees, say, is finite or infinite. So, in accordance with the definition, I take a subclass of the row of trees and investigate whether it is similar (i.e. can be co-ordinated one-to-one) to the whole class! (Here already the whole thing has become laughable.) It hasn't any meaning; for, ifI take a "finite class" as a sub-class, the attempt to coordinate it one-to-one with the whole class must eo ipso fail: and ifI make the attempt with an infinite class - but already that is a piece of nonsense, for if it is infinite, I cannot make an attempt to co-ordinate it. - What we call 'correlation of all the members of a class with others' in the case of a finite class is something quite different from what we, e.g., call a correlation of all cardinal numbers with all rational numbers. The two correlations, or what one means by these words in the two cases, belong to different logical types. An infinite class is not a class which contains more members than a finite one, in the ordinary sense of the word "more". If we say that an infinite number is greater than a finite one, that doesn't make the two comparable, because in that statement the word "greater" hasn't the same meaning as it has say in the proposition 5 > 4! That is to say, the definition pretends that whether a class is finite or infinite follows from the success or failure of the attempt to correlate a proper subclass with the whole class; whereas there just isn't any such decision procedure. - 'Infinite class' and 'finite class' are different logical categories; what can be significantly

asserted of the one category cannot be significantly asserted of the other. With regard to finite classes the proposition that a class is not similar to its sub-classes is not a truth but a tautology. It is the grammatical rules for the generality of the general implication in the proposition "k is a subclass of K" that contain what is said by the proposition that K is an infinite class. A proposition like "there is no last cardinal number" is offensive to naive - and correct - common sense. If I ask "Who was the last person in the procession?" and am told "There wasn't a last person" I don't know what to think; what does "There wasn't a last person" mean? Of course, if the question had been "Who was the standard bearer ?" I would have understood the answer "There wasn't a standard bearer" ; and of course the bewildering answer is modelled on an answer of that kind. That is, we feel, correctly, that where we can speak at all of a last one, there can't be "No last one". But of course that means: The proposition "There isn't a last one" should rather be: it makes no sense to speak of a "last cardinal number", that expression is ill-formed. "Does the procession have an end?" might also mean: is the procession a compact group? And now someone might say: "There, you see, you can easily imagine a case of something not having an end; so why can't there be other such cases?" - But the answer is: The "cases" in this sense of the word are grammatical cases, and it is they that determine the sense of the question. The question "Why can't there be other such cases?" is modelled on: "Why can't there be other minerals that shine in the dark"; but the latter is about cases where a statement is true, the former about cases that determine the sense. The form of expression "m = 2n correlates a class with one of its proper subclasses" uses a misleading analogy to clothe a trivial sense in a paradoxial form. (And instead of being ashamed of

I.

Cf. The Foundations of Arithmetic, §84. (Ed.)

this paradoxical form as something ridiculous, people plume themselves on a victory over all prejudices of the understanding). It is exactly as if one changed the rules of chess and said it had been shown that chess could also be played quite differently. Thus we first mistake the word "number" for a concept word like "apple", then we talk of a "number of numbers" and we don't see that in this expression we shouldn't use the same word "number" twice; and finally we regard it as a discovery that the number of the even numbers is equal to the number of the odd and even numbers. It is less misleading to say "m = 2ll allows the possiblity of correlating every time with another" than to say "m = 2n correlates all numbers with others". But here too the grammar of the meaning of the expression "possibility of correlation" has to be learnt. (It's almost unbelievable, the way in which a problem gets completely barricaded in by the misleading expressions which generation upon generation throw up for miles around it, so that it becomes virtually impossible to get at it.) If two arrows point in the same direction, isn't it in such a case absurd to call these directions equally long, because whatever lies in the direction of the one arrow, also lies in that of the other? The generality of m = 2n is an arrow that points along the series generated by the operation. And you can even say that the arrow points to infinity; but does that mean that there is somethinginfinity - at which it points, as at a thing? - It's as though the arrow designates the possibility of a position in its direction. But the word "possibility" is misleading, since someone will say: let what is possible now become actual. And in thinking this we always think of a temporal process, and infer from the fact that mathematics has nothing to do with time, that in its case possibility is already actuality. The "infinite series of cardinal numbers" or "the concept of cardinal number" is only such a possibility - as emerges clearly

from the symbol "I 0, ~, ~ + r I". This symbol is itself an arrow with the "0" as its tail and the "~+ r" as its tip. It is possible to speak of things which lie in the direction of the arrow, but misleading or absurd to speak of all possible positions for things lying in the direction of the arrow as an equivalent for the arrow itself. If a searchlight sends out light into infinite space it illuminates everything in its direction, but you can't say it illuminates infinity. It is always right to be extremely suspicious when proofs in mathematics are taken with greater generality than is warranted by the known application of the proof. This is always a case of the mistake that sees general concepts and particular cases in mathematics. In set theory we meet this suspect generality at every step. One always feels like saying "let's get down to brass tacks". These general considerations only make sense when we have a particular region of application in mind. In mathematics there isn't any such thing as a generalization whose application to particular cases is still unforseeable. That's why the general discussions of set theory (if they aren't viewed as calculi) always sound like empty chatter, and why we are always astounded when we are shown an application for them. We feel that what is going on isn't properly connected with real things. The distinction between the general truth that one can know, and the particular that one doesn't know, or between the known description of the object, and the object itself that one hasn't seen, is another example of something that has been taken over into logic from the physical description of the world. And that too is where we get the idea that our reason can recognize questions but not their answers.

i

I I

!,

Set theory attempts to grasp the infinite at a more general level than the investigation of the laws of the real numbers. It says that you can't grasp the actual infinite by means of mathematical symbolism at all and therefore it can only be described and not represented. The description would encompass it in something like the way in which you carry a number of things that you can't hold in your hand by packing them in a box. They are then invisible but we still know we are carrying them (so to speak, indirectly). One might say of this theory that it buys a pig in a poke. Let the infinite accommodate itself in this box as best it can. With this there goes too the idea that we can use language to describe logical forms. In a description of this sort the structures are presented in a package and so it does look as if one could speak of a structure without reproducing it in the proposition itself. Concepts which are packed up like this may, to be sure, be used, but our signs derive their meaning from definitions which package the concepts in this way; and if we follow up these definitions, the structures are uncovered again. (Cf. Russell's definition of "R*".) When "all apples" are spoken of, it isn't, so to speak, any concern of logic how many apples there are. With numbers it is different; logic is responsible for each and everyone of them. Mathematics consists entirely of calculations. In mathematics everything is algorithm and nothing is meaning; even when it doesn't look like that because we seem to be using words to talk abottt mathematical things. Even these words are used to construct an algorithm. In set theory what is calculus must be separated off from what attempts to be (and of course cannot be) theory. The rules of the game have to be separated off from inessential statements about the chessmen.

Frege replaced the signs with new words to show the definition wasn't really a definition. 1 Similarly in the whole of mathematics one might replace the usual words, especially the word "infinite" and its. cognates, with entirely new and hitherto meaningless exp:esslOns so as t~ see. what the calculus with these signs really ach1eves and what 1t falls to achieve. If the idea was widespread that chess gave us information about kings and castles I would propose to give the pieces new shapes and different nam:s, so as to demonstrate that everything belonging to chess has to be contained in the rules. . Wha~ a geome~rical proposition means, what kind of generality has, 1S somethmg that must show itself when we see how it is ~pplie~. For ~v~n if someone succeeded in meaning something ~ntang1ble ~y 1~ 1t ~ouldn't help him, because he can only apply it m a way wh1ch IS qU1te open and intelligible to everyone. Si~ila~ly, if so~eone imagi~ed the chess king as something mystical1t wouldn t worry us smce he can only move him on the 8 X 8 squares of the chess board. it

We have a feeling "There can't be possibility and actuality in mathematics. It's all on one level. And is in a certain sense, actttal. _ And that is correct. For mathematics is a calculus; and the calculus does not say of any sign that it is merely possible, but is concerned only wi~h the signs with which it actttal(y operates. (Compare the foundatlOns of set theory with the assumption of a possible calculus with infinite signs). When set theory appeals to the human impossibility of a direct sy~bolisation of the infinite it brings in the crudest imaginable ~s~nterpretation of its own calculus. It is of course this very m1s1nterpretation that is responsible for the invention of the calculus. But of course that doesn't show the calculus in itself to

In Cantor's alleged definition of "greater", "smaller", "+", "-" I.

Grundgesetze d. Arithmetik, II, § 83, pp. 93, 94.

41

be something incorrect (it would be at worst uninteresting) and it is odd to believe that this part of mathematics is imperilled by any kind of philosophical (or mathematical) investigations. (As well say that chess might be imperrilled by the dicovery that wars between two armies do not follow the same course as battles on the chessboard.) What set theory has to lose is rather the atmosphere of clouds of thought surrounding the bare calculus, the suggestion of an underlying imaginary symbolism, a symbolism which isn't employed in its calculus, the apparent description of which is really nonsense. (In mathematics anything can be imagined, except for a part of our calculus.)

The extensional conception of the real numbers

Like the enigma of time for Augustine, the enigma of the continuum arises because language misleads us into applying to it a picture that doesn't fit. Set theory preserves the inappropriate picture of something discontinuous, but makes statements about it that contradict the picture, under the impression that it is breaking with prejudices; whereas what should really have been done is to point out that the picture just doesn't fit, that it certainly can't be stretched without being torn, and that instead of it one can use a new picture in certain respects similar to the old one. The confusion in the concept of the "actual infinite" arises from the unclear concept of irrational number, that is, from the fact that logically very different things are called "irrational numbers" without any clear limits being given to the concept. The illusion that we have a firm concept rests on our belief that in signs of the the form "0. abcd ... ad infinitum" we have a pattern to which they (the irrational numbers) have to conform whatever happens. Ii

470

"Suppose I cut a length at a place where there is no rational point (no rational number)." But can you do that? What sort of a length are you speaking of? "But if my measuring instruments were fine enough, at least I could approximate without limit to a certain point by continued bisection"! - No, for I could never tell whether my point was a point of this kind. All I could tell would always be that I hadn't reached it. "But if! carry out the construction of V2 with absolutely exact drawing instruments, and then by bisection approximate to the point I get, I know that this process will never reach the constructed point." But it would be odd if one construction could as it were prescribe something to the others in this way! And indeed that isn't the way it is. It is very possible that

471

the point I get by means of the 'exact' construction of Vz is reached by the bisection after say 100 steps; - but in that case we could say: our space is not Euclidean.

it were to be added - how would it fill the gap? Suppose that it's 1t. If an irrational number is given through the totality of its approximations, then up to a'!Y point taken at random there is a series coinciding with that of 1t. Admittedly for each such series there is a point where they diverge. But this point can lie arbitrarily far 'out', so that for any series agreeing with 1t I can find one agreeing with it still further. And so if! have the totality of all irrational numbers except 1t, and now insert 1t I cannot cite a point at which 1t is now really needed. At every point it has a companion agreeing with it from the beginning on. To the question "how would we feel the lack of 1t" our answer must be "if 1t were an extension, we would never feel the lack of it". i.e. it would be impossible for us to observe a gap that it filled. But if someone asked us 'But have you then an infinite decimal expansion with the figure m in the r-th place and n in the s-th place, etc?' we could always oblige him.)

The "cut at the irrational point" is a picture, and a misleading picture. A cut is a principle of division into greater and smaller. Does a cut through a length determine in advance the results of all bisections meant to approach the point of the cut? No. In the previous example l in which I threw dice to guide me in the successive reduction of an interval by the bisection of a length I might just as well have thrown dice to guide me in the writing of a decimal. Thus the description "endless process of choosing between 1 and 0" does not determine a law in the writing of a decimal. Perhaps you feel like saying: the prescription for the endless choice between 0 and 1 in this case could be reproduced by a symbol like "0 ~~~ . . . ad. inf.". But if I adumbrate a law thus '0·001001001 . . . ad inf.", what I want to show is not the finite section of the series as a specimen of the infinite series, but rather the kind of regularity to be perceived in it. But in "0. ??? ... ad. inf." I don't perceive a'!Y law, - on the contrary, precisely that a law is absent. (What criterion is there for the irrational numbers being complete? Let us look at an irrational number: it runs through a series of rational approximations. When does it leave this series behind? Never. But then, the series also never comes to an end. Suppose we had the totality of all irrational numbers with one single exception. How would we feel the lack of this one? And - if

1.

I'

!

I

I

"The decimal fractions developed in accordance with a law still need supplementing by an infinite set of irregular infinite decimal fractions that would be 'brushed under the carpet' if we were to restrict ourselves to those generated l:Y a law." Where is there such an infinite decimal that is generated by no law? And how would we notice that it was missing? Where is the gap it is needed to fill ? What is it like if someone so to speak checks the various laws for the construction of binary fractions by means of the set of finite combinations of the numerals 0 and I? - The results of a law run through the finite combinations and hence the laws are complete as far as their extensions are concerned, once all the finite combinations have been gone through.

If one says: two laws are identical in the case where they yield the same result at every stage, this looks like a quite general rule. But in reality the proposition has different senses depending on

See below, p. 484

472

473

what is the criterion for their yielding the same result at every stage. (For of course there's no such thing as the supposed generally applicable method of infinite checking!) Thus under a mode of speaking derived from an analogy we conceal the most various meanings, and then believe that we have united the most various cases into a single system. (The laws corresponding to the irrational numbers all belong to the same type to the extent that they must all ultimately be recipes for the successive construction of decimal fractions. In a certain sense the common decimal notation gives rise to a common type.) We could also put it thus: elJe1Y point in a length can be approximated to by rational numbers by repeated bisection. There is no point that can only be approximated to by irrational steps of a specified type. Of course, that is only a way of clothing in different words the explanation that by irrational numbers we mean endless decimal fractions; and that explanation in turn is only a rough explanation of the decimal notation, plus perhaps an indication that we distinguish between laws that yield recurring decimals and laws that don't. The incorrect idea of the word "infinite" and of the role of "infinite expansion" in the arithmetic of the real numbers gives us the false notion that there is a uniform notation for irrational numbers (the notation of the infinite extension, e.g. of infinite decimal fractions). The proof that for every pair of cardinal numbers x and y Cy)2 =1= 2 does not correlate V2 with a single type of number - called "the irrational numbers". It is not as if this type of number was constructed before I construct it; in other words, I don't know any more about this new type of number than I tell myself.

474

42

Kinds of irrational numbers (n' P,F) n' is a rule for the formation of decimal fractions: the expansion of n' is the same as the expansion of n except where the sequence

P

F

777 occurs in the expansion of n; in that case instead of the sequence 777 there occurs the sequence 000. There is no method known to our calculus of discovering where we encounter such a sequence in the expansion of n. is a rule for the construction of binary fractions. At the nth place of the expansion there occurs a I or a 0 according to whether n is prime or not. is a rule for the construction of binary fractions. At the nth place there is a 0 unless a triple x, y, z from the first 100 cardinal numbers satisfies the equation xn + yn = zn.

I'm tempted to say, the individual digits of the expansion (of n for example) are always only the results, the bark of the fully grown tree. What counts, or what something new can still grow from, is the inside of the trunk, where the tree's vital energy is. Altering the surface doesn't change the tree at all. To change it, you have to penetrate the trunk which is still living. I call "nn" the expansion of n up to the nth place. Then I can say: I understand what n'lOO means, but not what n' means, since n has no places, and I can't substitute others for none. It would be 5--+3

different ifI e.g. defined the division alb as a rule for the formation of decimals by division and the replacements of every 5 in the quotient by a 3. In this case I am acquainted, for instance, with the 5--+3

number 1/7. - And if our calculus contains a method, a law, to calculate the position of 777 in the expansion of n, then the law of n includes a mention of 777 and the law can be altered by the substitution of 000 for 777. But in that case n' isn't the same as what

475

T I defined above; it has a different grammar from the one I supposed. In our calculus there is no question whether 7t :> 7t' or not, no such equation or inequality. 7t' is not comparable with 7t. And one can't say "notyet comparable", because if at some time I construct something similar to 7t' that is comparable to 7t, then for that very reason it will no longer be 7t'. For 7t' like 7t is a way of denoting a game, and I cannot say that draughts is notyet played with as many pieces as chess, on the grounds that it might develop into a game with 16 pieces. In that case it will no longer be what we call "draughts" (unless by this word I mean not a game, but a characteristic of several games or something similar; and this rider can be applied to 7t and 7t' too). But since being comparable with other numbers is a fundamental characteristic of a number, the question arises whether one is to call 7t' a number, and a real number; but whatever it is called the essential thing is that 7t' is not a number in the same sense as 7t. I can also call an interval a point and on occasion it may even be practical to do so; but does it become more like a point if I forget that I have used the word "point" with two different meanings? Here it is clear that the possibility of the decimal expansion does not make 7t' a number in the same sense as 7t. Of course the rule for this expansion is unambiguous, as unambiguous as that for 7t or v';-; but that is no proof that 7t' is a real number, if one takes comparability with rational numbers as an essential mark of real numbers. One can indeed abstract from the distinction between the rational and irrational numbers, but that does not make the distinction disappear. Of course, the fact that 7t' is an unambiguous rule for decimal fractions naturally signifies a similarity between n' and 7t or v';-; but equally an interval has a similarity with a point etc. All the errors that have been made in this chapter of the philosophy of mathematics are based on the confusion between internal properties of a form (a rule as one among a list of rules) and what we call "properties" in everyday life (red as a property

of this book). We might also say: the contradictions and unclarities are brought about by people using a single word, e. g. "number", to mean at one time a definite set of rules, and at another time a variable set, like meaning by "chess" on one occasion the definite game we play today, and on another occasion the substratum of a particular historical development. "How far must I expand 7t in order to have some acquaintance with it?" - Of course that is nonsense. We are already acquainted with it without expanding it at all. And in the same sense I might say that I am not acquainted with 7t' at all. Here it is quite clear that 7t' belongs to a different system from 7t; that is something we recognize if we keep our eyes on the nature of the laws instead of comparing "the expansions" of both. Two mathematical forms, of which one but not the other can be compared in my calculus with every rational number, are not numbers in the same sense of the word. The comparison of a number to a point on the number-line is valid only if we can say for every two numbers a and b whether a is to the right ofb or b to the right of a. It is not enough that someone should - supposedly - determine a point ever more closely by narrowing down its whereabouts. We must be able to construct it. To be sure, continued throwing of a die indefinitely restricts the possible whereabouts of a point, but it doesn't determine a point. After every throw (or every choice) the point is still infinitely indeterminate - or, more correctly, after every throw it is infinitely indeterminate. I think that we are here misled by the absolute size of the objects in our visual field; and on the other hand, by the ambiguity of the expression" to approach a point". We can say of a line in the visual field that by shrinking it is approximating more and more to a point - that is, it is becoming more and more similar to a point. On the other hand when a Euclidean line shrinks it does not become any more like a point; it always remains totally dissimilar, since its length, so to say,

477

never gets anywhere near a point. If we say of a Euclidean line that it is approximating to a point by shrinking, that only makes sense if there is an already designated point which its ends are approaching; it cannot mean that by shrinking it produces a point. To approach a point has two meanings: in one case it means to come spatially nearer to it, and in that case the point must already be there, because in this sense I cannot approach a man who doesn't exist; in the other case, it means "to become more like a point", as we say for instance that the apes as they developed approached the stage of being human, their development produced human beings. To say "two real numbers are identical if their expansions coincide in all places" only has sense in the case in which, by producing a method of establishing the coincidence, I have given a sense to the expression "to coincide in all places". And the same naturally holds for the proposition "they do not coincide if they disagree in atry one place". But conversely couldn't one treat 'It' as the original, and therefore as the first assumed point, and then be in doubt about the justification of 'It ? As far as concerns their extension, they are naturally on the same level; but what causes us to call 'It a point on the number-line is its comparability with the rational numbers. If I view 'It, or let's say Vl, as a rule for the construction of decimals, I can naturally produce a modification of this rule by saying that every 7 in the development of Vl is to be replaced by a 5; but this modification is of quite a different nature from one which is produced by an alteration of the radicant or the exponent of the radical sign or the like. For instance, in the modified law I am including a reference to the number system of the expansion which wasn't in the original rule for Vl. The alternation of the law is of a much more fundamental kind than might at first appear.

Of course, if we have the incorrect picture of the infinite extension before our_minds, it can appear as if appending the substitution rule 7-+5 to V 2 alters it much less than altering Vz into V;! because 7---",5

'

the expansions of Vl are very similar to those of Vl, whereas the expansion of V;! deviates from that of Vl from the second place onwards. Suppose I give a rule p for the formation of extensions in such a way that my calculus knows no way of predicting what is the maximum number of times an apparently recurring stretch of the extension can be repeated. That differs from a real number because in certain cases I can't compare p - a with a rational number, so that the expression p - a = b becomes nonsensical. If for instance the expansion of p so far known to me is 3· 14 followed by an open series of ones (3.141 I I I . . . ), it wouldn't be possible to say of the difference p -3· I 41 whether it was greater or less than 0; so in this sense it can't be compared with 0 or with a point on the number axis and it and p can't be called number in the same sense as one of these points. IThe extension of a concept of number, or of the concept 'all', etc. seems quite harmless to us; but it stops being harmless as soon as we forget that we have in fact changed our concept. I ISo far as concerns the irrational numbers, my investigation says only that it is incorrect (or misleading) to speak of irrational numbers in such a way as to contrast them with cardinal numbers and rational numbers as a different kind of number; because what are called "irrational numbers" are species of number that are really different - as different from each other as the rational numbers are different from each of them. I "Can God know all the places of the expansion of have been a good question for the schoolmen to ask.

'It?"

would

479

1 In these discussions we are always meeting something that could be called an "arithmetical experiment". Admittedly the data determine the result, but I can't see in what wcry they determine it. That is how it is with the occurrences of the 7S in the expansion of 71:; the primes likewise are yielded as the result of an experiment. I can ascertain that 3I is a prime number, but I do not see the connection between it (its position in the series of cardinal numbers) and the condition it satisfies. - But this perplexity is only the consequence of an incorrect expression. The connection that I think I do not see does not exist. There is not an - as it were irregular - occurrence of 7S in the expansion of 71:, because there isn't any series that is called the expansion of 71:. There are expansions of 71:, namely those that have been worked out (perhaps 1000) and in those the 7S don't occur "irregularly" because their occurrence can be described. (The same goes for the "distribution of the primes". If you give as a law for this distribution, you give us a new number series, new numbers.) (A law of the calculus that I do not know is not a law). (Only what I see is a law; not what I describe. That is the only thing standing in the way of my expressing more in my signs that I can understand.) Does it make no sense to say, even after Fermat's last theorem has been proved, that F = 0'1 I? (If, say I were to read about it in the papers.) I will indeed then say, "so now we can write 'F = o· I I'." That is, it is tempting to adopt the sign "F" from the earlier calculus, in which it didn't denote a rational number, into the new one and now to denote o· I I with it. F was supposed to be a number of which we did not know whether it was rational or irrational. Imagine a number, of which we do not know whether it is a cardinal number or a rational number. A description in the calculus is worth just as much as this particular set of words and it has nothing to do with an object given by description which may someday be found.

I

What I mean could also be expressed in the words: one cannot discover any connection between parts of mathematics or logic that was already there without one knowing. In mathematics there is no "not yet" and no "until further notice" (except in the sense in which we can say that we haven't yet multiplied two 1000 digit numbers together.) "Does the operation yield a rational number for instance ?" How can that be asked, if we have no method for deciding the question? For it is only in an established calculus that the operation ]ields results. I mean: "yields" is essentially timeless. It doesn't mean "yields, given time" - but: yields in accordance to the rules already known and established. "The position of all primes must somehow be predetermined. We work them out only successively, but they are all already determined. God, as it were, knows them all. And yet for all that it seems possible that they are not determined by a law." - Always this picture of the meaning of a word as a full box which is given us with its contents packed in it all ready for us to investigate. What do we know about the prime numbers? How is the concept of them given to us at all? Don't we ourselves make the decisions about them? And how odd that we assume that there must have been decisions taken about them that we haven't taken ourselves! But the mistake is understandable. For we use the expression "prime number" and it sounds similar to "cardinal number", "square number", "even number" etc. So we think it will be used in the same way, and we forget that for the expression "prime number" we have given quite different rules - rules different in kind - and we find ourselves at odds with ourselves in a strange way. - But how is that possible? After all the prime numbers are familiar cardinal numbers - how can one say that the concept of prime number is not a number concept in the same sense as the concept of cardinal number? But here again we are tricked by the image of an "infinite extension" as an analogue to the familiar "finite "extension. Of course the concept 'prime number'

is defined by means of the concept 'cardinal number', but "the prime numbers" aren't defined by means of "the cardinal numbers" and the way we derived the concept 'prime number' from th: concept 'cardinal number' is essentially different from that in which we derived, say, the concept 'square number'. (So we cannot be surprised if it behaves differently.) One might well imagine an arithmetic which - as it were - didn't stop at the concept 'cardinal number' but went straight on to that of square numbers. (Of course that arithmetic couldn't be applied in the same way as ours.) But then the concept "square number" wouldn't have the characteristic it has in our arithmetic of being essentially a part-concept, with the square numbers essentially a sub-class of the cardinal numbers; in that case the square numbers would be a complete series with a complete arithmetic. And now imagine the same done with the prime numbers! That will make it clear that they are not "numbers" in the same sense as e.g. the square numbers or the cardinal numbers. Could the calculations of an engineer yield the result that the strength of a machine part in proportion to regularly increasing loads must increase in accordance with the series of primes?

43

Irregular infinite decimals

"Irregular infinite decimals". We always have the idea that we only have to bring together the words of our everyday lang~age to give the combinations a sense, and all we then have to do 1S to inquire into it - supposing it's not quite clear right away.It is as if words were ingredients of a chemical compound, and we shook them together to make them combine with each other, and then had to investigate the properties of the compound. If someone said he didn't understand the expression "irregular infinite decimals" he would be told "that's not true, you understand it very well: don't you know what the words "irregular", "infinite", and "decimal" mean? - well, then, you understand their combination as well." And what is meant by "understanding" here is that he knows how to apply these words in certain cases, and say connects an image with them. In fact, someone who puts these words together and asks "what does it mean" is behaving rather like small children who cover a paper with random scribblings, show it to grown-ups, and ask "what is that?" "Infinitely complicated law", "infinitely complicated construction" ("Human beings believe, if only they hear words, there must be something that can be thought with them"). How does an infinitely complicated law differ from the lack of any law? (Let us not forget: mathematicians' discussions of the infinite are clearly finite discussions. By which I mean, they come to an end.) "One can imagine an irregular infinite decimal being constr~cted by endless dicing, with the number of pips in each case bemg a decimal place." But, if the dicing goes on for ever, no final result ever comes out.

"It is only the human intellect that is incapable of grasping it, a higher intellect could do so!" Fine, then describe to me the grammar of the expression "higher intellect"; what can such an intellect grasp and what can't it grasp and in what cases (in experience) do I say that an intellect grasps something? You will then see that describing grasping is itself grasping. (Compare: the solution of a mathematical problem.) Suppose we throw a coin heads and tails and divide an interval AB in accordance with the following rule: "Heads" means: take the left half and divide it in the way the next throw prescribes. "Tails" says "take the right half, etc." By repeated throws I then

A

I

B

I II/I

I

get dividing-points that move in an ever smaller interval. Does it amount to a description of the position of a point if I say that it is the one infinitely approached by the cuts as prescribed by the repeated tossing of the coin? Here one believes oneself to have determined a point corresponding to an irregular infinite decimal. But the description doesn't determine a'!)' point explicitlY; unless one says that the words "point on this line" also "determine a point"! Here we are confusing the recipe for throwing with a mathematical rule like that for producing decimal places of Yz. Those mathematical rules are the points. That is, you can find relations between those rules that resemble in their grammar the relations "larger" and "smaller" between two lengths, and that.is why they are referred to by these words. The rule for working out places of v'2 is itself the numeral for the irrational number; and the reason I here speak of a "number" is that I can calculate with these signs (certain rules for the construction of rational numbers) just as I can with rational numbers themselves. If I want to say similarly

that the recipe for endless bisection according to heads and tails determines a point, that would have to mean that this recipe could be used as a numeral, i.e. in the same way as other numerals. But of course that is not the case. If the recipe were to correspond to a numeral at all, it would at best correspond to the indeterminate numeral "some", for all it does is to leave a number open. In a word, it corresponds to nothing except the original interval.

Note in Editing

In June 193 I Wittgenstein wrote a parenthesis in his manuscript book: "(My book might be called: Philosophical Grammar. This title would no doubt have the smell of a textbook title but that doesn't matter, for behind it there is the book.)" In the next four manuscript volumes after this he wrote nearl y everything that is in the present work. The second of these he callc:d "Remarks towards Philosophical Grammar" and the last two "Philosophical Grammar". The most important source for our text is a large typescript completed probably in 1933, perhaps some of it 1932. Our "Part II" makes up roughly the second half of this typescript. In most of the first half of it Wittgenstein made repeated changes and revisions - between the lines and on the reverse sides of the typed sheets - and probably in the summer of 1933 he began a "Revision" in a manuscript volume (X and going over into XI). This, with the "Second Revision" (which I will explain), is the text of our Part I up to the Appendix. - Wittgenstein simply wrote "Umarbeitung" (Revision) as a heading, without a date; but he clearly wrote it in 1933 and the early weeks of 1934. He did not write the "second revision" in the manuscript volume but on large folio sheets. He He crossed out the text that this was to replace, and showed in margins which parts went where. But it is a revision of only a part, towards the beginning, of the first and principal "Revision". The passages from the second revision are, in our text, §§1-1 3 and §§2343. The second revision is not dated either, but obviously it is later than the passages it replaces; probably not later than 1934. So we may take it that he wrote part of this work somewhat earlier, and part at the same time as his dictation of The Blue Book. Many things in the Blue Book are here (and they are better expressed). There are passages also which are in the Philosophical Remarks and others later included in the Investigations. It would be easy to give the reference and page number for each of these. We decided not to. This book should be compared with Wittgenstein's earlier and later writings. But this means: the method and the development

of his discussion here should be compared with the Ph':losophical Remarks and again with the Investigations. The footnotes would be a hindrance and, as often as not, misleading. When Wittgenstein writes a paragraph here that is also in the Remarks, this does not mean that he is just repeating what he said there. The paragraph may have a different importance, it may belong to the discussion in a different way. (We know there is more to be said on this question.) Wittgenstein refers to "my book" at various times in his manuscripts from the start of 1929 until the latest passages of the Investigations. It is what his writing was to produce. The first attempt to form the material into a book was the typescript volume he made in the summer of 1930 - the Philosophical Remarks (published in German in 1964). The large typescript of 1933 - the one we mentioned as a source of this volume - looks like a book. Everyone who sees it first thinks it is. But it is unfinished; in a great many ways. And Wittgenstein evidently looked on it as one stage in the ordering of his material. (Cf. the simile of arranging books on the shelves of a library, in Blue Book p. 44-45.) Most of the passages which make up the text of the 1933 typescript (called" 2 1 3" in the catalogue) he had written in manuscript volumes between July 1930 and July 1932; but not in the order they have in the typescript. From the manuscript volumes he dictated two typescripts, one fairly short and the other much longer - about 8 50 pages together. There was already a typescript made from manuscripts written before July 1930 - not the typescript which was the Philosophical Remarks but a typescript which he cut into parts and sifted and put together in a different way to make the Philosophical Remarks. He now used an intact copy of this typescript together with the two later ones in the same way, cutting them into strips: small strips sometimes with just one paragraph or one sentence, sometimes groups of paragraphs; and arranging them in the order he saw they ought to have. Groups of slips in their order were clipped together to form 'chapters', and he gave each chapter a title. He then brought the chapters togetherin a definite order - to form 'sections'. He gave each section a title and arranged them also in a definite order. In this order the whole was finally typed. - Later he made a table of contents out of the titles of sections and chapter headings.

Certain chapters, especially, leave one feeling that he cannot have thought the typing of the consecutive copy had finished the work barring clerical details. He now wrote, over and over again, between the lines of typescript or in the margin: "Does not belong here", "Belongs on page ... above", "Belongs to 'Meaning', § 9", "Goes with 'What is an empirical proposition?' ", "Belongs with §14, p. 58 or § 89 p. 414", and so on. But more than this, about 350 pages - most of the first half of the typescript - are so written over with changes, additions, cancellations, questions and new versions, that no one could ever find the 'correct' text here and copy it - saving the author himself should write it over to include newer versions and make everything shorter. He now makes no division into chapters and sections. He has left out paragraph numbers and any suggestion of a table of contents. We do not know why. (We do not find chapters or tables of contents anywhere else in Wittgenstein's writings. He may have found disadvantages in the experiment he tried here.) - The extra spaces between paragraphs and groups of paragraphs are his own; and he thought these important. He would have numbered paragraphs, probably, as he did in the Investigations. But the numbers in Part I here are the editor's, not Wittgenstein's. Neither is the division in chapters Wittgenstein's, nor the table of contents. - On the other hand, Part II has kept the chapters and the table of contents which Wittgenstein gave this part of the typescript. Perhaps this makes it look as though Part I and Part II were not one work. But we could not make them uniform in this (division and arrangement of chapters) without moving away from Wittgenstein's way of presenting what he wrote. Anyone who reads both parts will see connections. And the appendix may make it plainer. Appendices 5, 6, 7, 8 and the first half of 4 are chapters of 'typescript 213'. Appendix 1, Fact and Complex, is also an appendix in Philosophical Remarks. But Wittgenstein had fastened it together with appendices 2 and 3 and given them a consecutive paging as one essay; with what intention we do not know. Each one of the eight appendices here discusses something connected with 'proposition' and with 'sense of a proposition'. The whole standpoint is somewhat earlier (the manuscripts often bear earlier dates) than that of

Part I here, but later than the Philosophical Remarks. - But the appendices also discuss questions directly connected with the themes of 'generality' and 'logical inference' in Part II. Part I is concerned with the generality of certain expressions or concepts, such as 'language', 'proposition' and 'number'. For instance, § 70, page I I 3 : "Compare the concept of p.roposition with the concept 'number' and then on the other hand with the concept of cardinal number. We count as numbers cardinal numbers rational numbers irrational numbers, complex numbers; whether we call othe; constructions numbers because of their similarities with these or draw a definitive boundary here or elsewhere, depends on' us. In this respect the concept of number is like the concept of proposition. On the other hand the concept of cardinal number II, ~, ~ + I I can be called a rigorously circumscribed concept, that's to say it's a concept in a different sense of the word." This discussion is closely related to the chapter on 'Kinds of Cardinal Numbers' and on '2 + 2 = 4' in Part II; and with the section on Inductive Proof. These are the most important things in Part II.

Translator's Note

Many passages in the Philosophical Grammar appear also in the Philosophical Remarks, the Philosophical Investigations, and the Zettel. In these cases I have used the translations of Mr Roger White and Professor G. E. M. Anscombe, so that variations between the styles of translators should not be mistaken for changes of mind on Wittgenstein's part. Rare departures from this practice are marked in footnotes. Passages from the Philosophical Grammar appear also in The Principles of Linguistic Philosop~J' of F. Waismann (Macmillan 1965): in these cases I have not felt obliged to follow the English text verbatim, but I am indebted to Waismann's translator. Three words or groups of words constantly presented difficulties in translation. The German word "Satz" may be translated "proposition" or "sentence" or (in mathematical and logical contexts) "theorem". I have tried to follow what appears to have been Wittgenstein's own practice when writing English, by using the word "proposition" when the syntactical or semantic properties of sentences were in question, and the word "sentence" when it was a matter of the physical properties of sounds or marks. But it would be idle to pretend that this rule provides a clear decision in every case, and sometimes I have been obliged to draw attention in footnotes to problems presented by the German word. From the Tractatus onward Wittgenstein frequently compared a proposition to a MaJ3stab. The German word means a rule ~r measuring rod: when Wittgenstein used it is clear that he had 1U mind a rigid object with calibrations. Finding the word "rule" too ambiguous, and the word "measuring-rod" too cumbersome, I have followed the translators of the Tractatus in using the less accurate but more natural word "ruler". Translators of Wittgenstein have been criticised for failing to adopt a uniform translation of the word "iibersehen" and its derivatives, given the importance of the notion of "iibersichtliche Darstellung" in Wittgenstein's later conception of philosophy. I have

49 1

been unable to find a natural word to meet the requirement of uniformity, and have translated the word and its cognates as seemed natural in each context. Like other translators of Wittgenstein I have been forced to retain a rather Germanic style of punctuation to avoid departing too far from the original. For instance, Wittgenstein often introduced oratio recta by a colon instead of by inverted commas. This is not natural in English, but to change to inverted commas would involve making a decision - often a disputable one - about where the quotation is intended to end. I have translated the text of the Suhrkamp-Blackwell edition of 19 69 as it stands, with the exception of the passages listed below in which I took the opportunity to correct in translation errors of transcription or printing which had crept into the German text. The pagination of the translation, so far as practicable, matches that of the original edition. I am greatly indebted to Professor Ernst Tugendhat, who assisted me in the first draft of my translation; and to Mr John Thomas, Dr Peter Hacker, Mr Brian McGuinness, Professor G. E. M. Anscombe, Professor Norman Malcolm, Professor G. H. von Wright, Mr Roger White, Dr Anselm Miiller, Mr and Mrs J. Tiles and Mr R. Heinaman who assisted me on particular points. My greatest debt is to Mr Rush Rhees, who went very carefully through large sections of a draft version and saved me from many errors while improving the translation in many ways. The responsibility for remaining errors is entirely mine. I am grateful to the British Academy for a Visiting Fellowship which supported me while writing the first draft translation. Oxford 1973

Corrections to the Ig6g German Edition

page

line 31 For "Gedanken" read "Gedanke". 24 15 For "selten" read "seltsam". 25 17 For "Vom Befehl" read "Von der Erwartung". For "Carroll's" read "Carroll's Gedicht". 23 43 For "was besagt" read "was sagt". 15 44 For "uns da" read "UflS da etwas". 13 52 27 For "geben" read "ergeben". 52 61 22 For "kann" read "kann nun". 2 4 There should be no space between the para7 graphs. For "iibergrenzt" read "iiberkreuzt". 28 75 For "gezeichnet haben" read "bezeichnet 88 9 haben, sie schon zur Taufe gehalten haben". For "nun" read "nun nicht". 2 27 9 7 For "Jedem" read "jedem". 97 For "sie" read "sie von". 107 108 17 For "Korperlos" read "korperlos". 8 For "nun" read "urn". 119 12 3 16 There should be no space between the paragraphs. For "daB" read "das". 21 147 For "Zeichen" read "Zeichnen". 148 21 For "schreibt" read "beschreibt". 151 15 2 19 For "Auszahlungen" read "Auszahnungen". en rea d "hervorzuru£en " . 152 30 For" h ervorgeru£" 160 21 There should be a space between the paragraphs. 160 15 For "Ciffre" read "Chiffre". 16 3 30 There should be no space between the paragraphs. . h en " rea d" sc hema t"ISCh en " . For "systematlsc 1 12 17 For "liigen:" read "liigen"". 8 17 3 For "in der" read "der". 20 5 9 21 5 3 For "Buches" read "Buches uns". 221 8 For "Erkenntis" read "Erkenntnis". 493 17

page 244 line 20 For "folgen" read "Folgen". 244 244 245 251 251

23 24 19 12 I3

253 25 6 260 261 268 27 8

23 23 I2 Iff 3I

6

For "etwa" read "es dazu". For "den" read "dem". For "fy" read "fx". For "dem" read "den". For "mit" read "mir". The "tI" in the figure is misplaced. For "folge" read "folgte". For "weh" read "nicht weh". For "ist" read "ist da". For 'f' read '~' passim. For "Fall von f(3) ist," read "Fall von f(3) ist. Und nun kann mans uns entgegenhalten: Wenn er sieht, dass f(a) ein Fall von f(3) ist," For "0'3" read "0'3". For "1+1+1+1" read "1+1+1+1

286 288

II

325

30 For "\)ix" read "~x". 5 For "wird" read "wird. Yom Kind nur die richtige Ausfiihrung der Multiplikation verlangt wird". 13 For "uns" read "nun uns". I 2 For "Ich habe gesagt" read "Ich sagte oben". 12 Insert new paragraph: "Wenn nachtraglich ein Widerspruch gefunden wird, so waren vorher die Regeln noch nicht klar und eindeutig. Der Widerspruch macht also nichts denn er ist dann durch das Aussprechen einer Regel zu entfernen." 8 For "Hiweisen" read "Hinweisen". 21 For "unnotiges Zeichen fiir "Taut." geben" read "unnotiges - Zeichen fiir "Taut." gegeben". 6 For "den" read "dem". For "Def." read "~~r" For "Satze)." read "Satze) und zwar eine richtige degenerierte Gleichung (den Grenzfall einer Gleichung)." 21 For "konne" read "konnte".

+

494

I".

line 27 For "nenne" read "nennen". 9 For "Schma" read "Schema". For "(x" read "(E x". 344 9 For "(x)" read "(E x)". 344 2 For "Cont." read "Kont." 353 13 For "(3)" read "(3n) 353 22 For "nur" read "nun". 38 3 86 13 For "3n" read "3n2". 3 II For "the" read "der". 388 For "keine" read "eine". 393 For "Schreibmaschine" read "Schreibweise". 393 For "I: 2" read "I: 2". 8 39 to tenth ~ords of title should be Seventh 0 4 5 roman. For "die Kardinalzahlen" read "aIle Kardinalzahlen" . fn2 For "--+" read "S.4I4". 27 For "konnen." read "konnen. Niemand aber wiirde sie in dies em Spiel einen Beweis genannt haben !" For "(b + I" read "(b + I)". 16 tn4 4 20 For "((4 + I) + I" read "((4 + I) + I)". 1 43 16 For "gesehn" read "gesehen". 441 441 33 For "pll" read "~". 4 For "3n + 7" read "3 + n = 7". 451 453 32 F or "( .. . )" read "( . .. )" . 6 45 35 For "Kann nicht" read "kann". 60 4 7 For "konnten?" read "konnten? Unser Netz ware also nicht fein genug ?" 13 For "listen" read "Listen". 28 For "andere Mineralien" read "auch andere Falle". For "erhohten" read "erhaltenen". The symbol should read: "0, ??? ... ". For "Quadratzahlen", gerade Zahlen" read " "Quadratzehlen", "gerade Zahlen". For "primzahl" read "Primzahl".

page 325 328

~" ~ OSTREDNA :,..

I

.g

Y.lllliNlCA -

495